Global Debt Report 2026

Sustaining Debt Market Resilience Under Growing Pressure

Report

4 March 2026

2. Corporate debt market outlook in a transforming world

Copy link to 2. Corporate debt market outlook in a transforming world

Abstract

The global economy faces great uncertainty. Macroeconomic and geopolitical headwinds have raised the prospect of downward pressure on economic growth, with elevated volatility in many markets, including traditionally safe assets. Despite this context, corporate credit spreads are near historical lows even for riskier debtors, and corporate borrowing reached record levels in 2025, with enormous capital needs for the artificial intelligence expansion set to take it even higher. This chapter explores the drivers of these trends and what their implications will be.

Key findings

Copy link to Key findings

Corporate debt markets are simultaneously subject to – and themselves drivers of – two transformative contemporary developments: the advancement of artificial intelligence and the ongoing shifts in the global economic and political landscape.
Global corporate debt issuance reached about USD 13.7 trillion in 2025 (USD 6.8 trillion in corporate bonds and USD 7 trillion in syndicated loans), the highest amount on record. Outstanding amounts stood at USD 59.5 trillion at the end of the year, comprising USD 36.4 trillion of corporate bonds and USD 23.1 trillion of syndicated loans.
Corporate credit spreads globally are near historical lows for both investment grade and non-investment grade companies, despite geopolitical tensions, possible future macroeconomic headwinds and record levels of borrowing.
This aligns with strong fundamentals in the corporate sector: cash levels and forecasted earnings are high, aggregate credit quality indicators are solid, and projected default rates are below historical averages.
Several factors unrelated to corporate credit quality are also putting downward pressure on spreads. Notably, the world has seen a shift in relative risk from corporate to sovereign debt markets, increasing the benchmark against which corporate borrowing costs are measured. As a consequence, some major index companies are even trading at negative spreads to their government benchmark.
A decomposition analysis suggests that a large part of credit spread reductions in recent years comes from reductions in liquidity premia. This likely stems in part from changes in the investor base – notably the increasing presence of investment funds and ETFs and, more recently, principal trading firms – and developments in corporate bond market trading structures. These trends are reinforcing one another, as greater liquidity attracts more trading-prone investors. Together with increases in investor risk-willingness, this means corporate spreads now primarily reflect compensation for expected default losses.
The AI expansion is immensely capital intensive. The projected capital expenditure for nine major players, often called hyperscalers (primarily US technology firms), amounts to USD 4.1 trillion between 2026 and 2030. For comparison, total capital expenditure by all non-financial companies in the United States in 2025 was just above USD 3 trillion. If half of the future capex needs are financed through bond markets, borrowing by these nine issuers alone would amount to an average of 15% of historical global gross issuance per year.
These two developments – AI spending dominating global markets and changes in trading frequency – beg the question of whether debt markets are becoming more like equity markets. The technology sector represents an increasing share of the global bond market, a trend that is set to continue given the magnitude of the hyperscalers’ financing needs. This might bring bond market sector, and even single‑firm, concentration closer to that observed in equity markets in recent years. Given the uncertainty about the useful life of data centres, a key AI infrastructure, and the nature of their value as collateral, there seems to be a convergence even in the type of risks financed by equity and debt markets.
Corporate equity and debt market pricing is also increasingly correlated. The co-movement of credit spreads and equity prices (hedge ratios) has increased markedly in recent years for major indices. The change in investor base may have contributed to this increase. It may also be an effect of increasing constituent convergence between major bond and equity indices.

This chapter looks at developments in corporate borrowing across bond, syndicated loan, and private credit markets for both financial and non-financial companies, drawing from original OECD datasets.

Corporate debt markets in a transforming world

Copy link to Corporate debt markets in a transforming world

Two simultaneous developments are transforming the world economy: the exponential advance in artificial intelligence (AI) and increasing global economic and political uncertainty. Both will shape debt markets in the years to come – but equally, debt markets will play a part in determining the outcome of this transformation.

The financing needs for the AI expansion are unprecedented in recent history. The magnitude of the capital expenditure needed to build the infrastructure required to operate at scale is so great that increases in external financing are inevitable. This applies not just to the technology sector, but to the entire supply chain of industries involved, from energy providers to construction companies. Debt markets will be used to mobilise a very significant part of that financing – but corporate debt levels are already elevated, and with continued fiscal pressures on governments and a changing investor base (see Chapters 1 and 3), the question is how much more debt can be absorbed in an orderly manner. In a positive scenario, AI technology might lead to substantial productivity gains, mitigating concerns about broad-market debt sustainability. But that outcome is not guaranteed; while AI itself holds great promise for increased future prosperity, identifying what corporations will end up reaping the greatest benefits is an open question. The useful life of underlying assets might change rapidly given new developments. In this sense, certain current corporate bets on AI have a risk profile that looks more equity than debt-like – but with the nominal repayment requirements of a debt contract.

In parallel, the global trade order that has prevailed for the last decades is experiencing major changes, with the possibility of being redrawn altogether. Geopolitical tensions and structural fiscal pressures, among other challenges, add additional strain on markets. This not only puts potential downside pressure on economic growth prospects but might also reshape international investment flows. Consequently, indicators of economic and financial uncertainty have been significantly elevated in recent years.

Corporate debt markets have been remarkably unfazed in the face of these challenges. Credit spreads are near historical lows globally, including for riskier non-investment grade debt, even as borrowing reached a new peak in 2025 (Figure 2.1). This chapter looks at the drivers of what at first glance appears to be a dislocation between market dynamics and broader political and economic uncertainty. It seeks to tie ongoing trends in corporate debt markets to broader global developments and analyse what the implications of these might be for market functioning, debt sustainability – and the economy at large.

Figure 2.1. Corporate bond spreads and economic uncertainty
Copy link to Figure 2.1. Corporate bond spreads and economic uncertainty

Corporate bond spreads are near historical lows despite elevated policy and macroeconomic uncertainty

Headline developments in global corporate debt markets

Copy link to Headline developments in global corporate debt markets

Corporate debt issuance in 2025 was the highest ever. Companies globally borrowed USD 13.7 trillion across corporate bond and syndicated loan markets, surpassing the previous peak in 2021, when debt markets played a central role in helping firms meet pandemic-related financing needs, fuelled by very expansionary monetary policy. The increase is a continuation of the 2024 trend towards growing borrowing (notably by non-financial companies), which marked a sharp reversal from the decline seen in 2022 and 2023, when volumes fell to some of the lowest levels of the past decade in response to higher interest rates (Figure 2.2, Panel A). Despite record levels of gross issuance, outstanding debt levels have not grown markedly since 2024, remaining broadly stable at around USD 60 trillion, still lower in real terms than the 2020 peak (Panel B).

Figure 2.2. Global corporate debt market activity
Copy link to Figure 2.2. Global corporate debt market activity

At 13.7 trillion, corporate gross borrowing from markets in 2025 was the highest on record in real terms

Total private credit assets under management (AUM) have also remained broadly stable, standing at USD 1.8 trillion in June 2025, 1% more than in 2024 (Figure 2.3). This represents a 2% reduction in real terms, compared to a compound average real growth rate of 14% from 2000-2024. The ratio of undeployed committed capital (“dry powder”) to total assets under management fell to 28% in 2025, continuing a downward trajectory that started in 2018 when it stood at 59%, indicating that the period of breakneck-paced capital raising has given way to a period of allocating those funds to actual investments. As highlighted later in this chapter, private credit markets are set to play, and are indeed already playing, an important role in financing the global AI expansion.

Figure 2.3. Global private credit assets under management by investment status
Copy link to Figure 2.3. Global private credit assets under management by investment status

Private credit assets have stabilised after years of very rapid growth, with the share of dry powder decreasing

Net issuance (gross issuance minus redemptions) of corporate bonds continued to expand globally in 2025 (Figure 2.4). Investment grade companies added USD 768 billion of new debt, split in roughly equal parts between financial and non-financial companies (Panel A). In the riskier, non-investment grade segment, net issuance turned positive again for non-financial companies for the first time since 2021 (Panel B).

Figure 2.4. Net issuance of corporate bonds
Copy link to Figure 2.4. Net issuance of corporate bonds

Net issuance continues to expand for IG companies and has turned positive for higher risk non-financial firms

As noted in the 2025 edition of the Global Debt Report, despite elevated interest rates since 2022, there was still a gap between the cost of outstanding debt and the interest paid by companies when issuing new debt at the end of 2024, nearly three years after the hiking cycle began. Owing to long-term maturities and predominantly fixed-rate debt, the cost of the outstanding debt stock did not yet reflect the new financing environment. Globally, in 2024 investment grade companies paid roughly 1 percentage point (p.p.) more for new debt than the cost of their outstanding debt; for non-investment grade issuers the difference was 1.4 p.p. (OECD, 2025[2]). That gap has since narrowed substantially, roughly halving for both risk categories, driven both by the cost of the outstanding stock of debt slowly increasing as old debt is refinanced and falling interest costs at issuance as monetary policy became more accommodative (Figure 2.5).

Figure 2.5. Interest cost at issuance versus effective cost of outstanding bond debt, global
Copy link to Figure 2.5. Interest cost at issuance versus effective cost of outstanding bond debt, global

The cost of outstanding debt and the cost of issuing new debt are converging

This increase in the cost of debt is also starting to reflect in the outstanding debt stock (Figure 2.6). Given the dynamics described above, the cost of outstanding debt is very slow-moving, but the structural trend towards higher interest rates is now clearly visible. For investment grade companies, half of outstanding debt now carries an interest cost above 4%, the first time since 2015. The share of ultra-cheap debt, at rates of 2% or less, now represents 14% of outstanding amounts, down from almost a quarter in 2021 (Panel A). The adjustment has been somewhat faster for non-investment grade debt, given shorter average maturities in this riskier segment. At the end of 2025, 15% of debt cost 8% or more, up from 9% of outstanding bonds in 2022 (Panel B).

Figure 2.6. Outstanding global corporate bond debt by interest rate (coupon)
Copy link to Figure 2.6. Outstanding global corporate bond debt by interest rate (coupon)

Half of outstanding investment grade debt now has an interest rate above 4%

This trend is set to continue in the coming years. Refinancing requirements in the next three years amount to 24% of outstanding investment grade debt and 31% of non-investment grade debt (as of the end of 2025). The majority of debt maturing has a lower coupon than the current cost of outstanding debt; for investment grade companies, 65% of debt due between 2026 and 2028 has an interest rate of 4% or less. Similarly, 67% of non-investment grade debt coming due in the same period currently costs 6% or less (Figure 2.7).

Figure 2.7. Refinancing requirements in the next five years by cost of outstanding debt
Copy link to Figure 2.7. Refinancing requirements in the next five years by cost of outstanding debt

Debt maturing in the near term is predominantly legacy borrowing that will need to be refinanced at a higher cost

Interpreting compressed credit spreads in an otherwise volatile environment

Copy link to Interpreting compressed credit spreads in an otherwise volatile environment

In addition to other headwinds, both geopolitical and macroeconomic, in the last four years corporate bond markets were also exposed to the fastest monetary policy tightening cycle in recent history. While this global tightening, which started in 2022, did see credit spreads move upward, the increase was not proportionate to the pace of the rate hikes, and the prospect of a macroeconomic “soft landing” that took hold relatively shortly after rapidly saw spreads compress again.

The muted spread response to tighter monetary policy is not necessarily unexpected; there is no mechanical relationship between policy rates and corporate credit spreads, with the movement of the latter depending heavily on the drivers of changes in the former. The observed rate‑spread correlation will be very different depending on whether rate hikes are a response to higher growth or to “growth-independent” inflationary episodes. In the former case, there is an economic logic to credit spreads remaining unchanged or even tightening. This dynamic of gradually increasing policy rates and compressed spreads was seen in the United States in the 2004-2006 period (Figure 2.8).

Even so, there are conceptual reasons to expect higher rates to be associated with higher spreads. Higher policy rates serve to lower aggregate demand and tighten financial conditions, thereby, all else equal, increasing credit risk premia through higher default risk. There is also the possibility of a liquidity effect whereby higher rates lead to lower liquidity in corporate securities as trading concentrates in government debt markets, increasing liquidity premia and therefore spreads. In addition, rate increases during the cycle that began in 2022 were not primarily driven by concerns about an overly hot economy. Inflation was an effect of other developments, notably (but by no means exclusively) spikes in energy prices and shipping costs. In spite of this, spreads on both investment grade and non-investment grade corporate debt have remained subdued.

Figure 2.8. Corporate bond spreads and US monetary policy rate
Copy link to Figure 2.8. Corporate bond spreads and US monetary policy rate

Spreads remained compressed even as monetary policy was tightened at the most rapid pace in recent history

What explains the surface‑level calm of corporate credit markets in the face of what appears to be major economic headwinds? For analytical purposes, the driving factors can usefully be split into two broad categories: fundamental and technical. Fundamental factors are those affecting credit risk, i.e. developments that have improved corporate credit quality relative to sovereign debt, such as increased cash buffers, reduced leverage or higher earnings prospects. Technical factors, on the other hand, are those that influence spreads but that are not necessarily indicative of improved credit quality (e.g. exogenous shifts in demand/supply or changes in liquidity).

Corporate credit quality

Developments in credit quality indicators are directionally aligned with increasingly compressed spreads. Corporate cash levels are ample, remaining above historical averages despite decreasing since the spike during the COVID‑19 pandemic. The increase is particularly visible among US companies, although less markedly when adjusting for the so-called hyperscalers’ substantial cash holdings (Figure 2.9, Panel A). Corporate earnings prospects are also strong. Earnings forecasts (one‑year forward EPS) for companies representative of the investment grade index in Figure 2.8 were 12% higher in 2025 than they were in 2013 (Panel B) and the median interest coverage ratio was 6.9 in 2024 (latest available data), compared to a historical average of 5.9 (Panel C). Default rates are trending downwards globally and are forecast to fall substantially below historical averages going forward (Panel D). These developments, especially in an environment of increasing sovereign risk, are all consistent with a compression of corporate spreads.

Figure 2.9. Measures of corporate sector financial strength
Copy link to Figure 2.9. Measures of corporate sector financial strength

Fundamental indicators of corporate sector health and financial strength are aligned with compressed spreads

Other market-wide indicators also suggest an improvement in risk profile, both when it comes to credit risk and interest rate risk (Figure 2.10). The value‑weighted credit ratings of both investment grade and non-investment grade bond issues have seen an upward movement, albeit minor (Panel A). Sensitivity to interest rate changes, measured as duration, is down significantly, in particular for investment grade companies (Panel B).

Figure 2.10. Credit ratings and index duration
Copy link to Figure 2.10. Credit ratings and index duration

The credit quality of newly issued bonds inched upwards in 2025 and interest rate risk has decreased

Shift in relative risk between corporate and sovereign debt markets

Yet improvements in credit quality alone do not explain the current tightness of credit spreads. Several other factors are also at play. One key consideration is the relative shift in risk from corporate to sovereign markets. Since the 2008 financial crisis, government borrowing has outpaced corporate borrowing (see Chapter 1 and BIS (2025[3])). The response to the global financial crisis and later COVID‑19 pandemic led to spikes in sovereign debt that did not revert to pre‑crisis levels after the crises, resulting in a seemingly permanent situation of debt at levels unprecedented in peacetime. The rise in interest rates to combat inflation in 2022 put significant upward pressure on government yields, and thereby on government interest expenses, which exceeded major government spending areas such as defence in the OECD on aggregate in 2024 (OECD, 2025[2]). This increase in the cost of debt has in turn further increased concerns about fiscal sustainability, which in combination with a larger supply to be absorbed is putting additional upward pressure on borrowing costs. This is all exacerbated by an environment of elevated geopolitical and macroeconomic uncertainty as well as the reduction of substantial central bank holdings as quantitative easing programmes are unwound (see Chapters 1 and 3). This helps explain why government yields have not fallen in tandem with interest rates since the monetary policy hiking cycle ended (Figure 2.11, Panel A).

Taken together, the increasing risk associated with government debt and improving corporate fundamentals in the publicly traded market have meant that corporate yields did not increase in proportion to sovereign yields, the result being a compression in spreads between the two. This is also reflected by a record number of companies with bonds in the global reference index used in this chapter (see Box 2.1) trading at negative spreads against their government benchmark securities (Panels B and C). At 1.6% of index companies at the end of 2025, negative spreads remain an exception, but the recent increase is stark.

Figure 2.11. Benchmark yields and corporate bonds with negative spreads
Copy link to Figure 2.11. Benchmark yields and corporate bonds with negative spreads

As benchmark yields have increased, some companies are even trading at negative spreads to their sovereign equivalents

The changing structure and liquidity of corporate debt markets

A number of more technical factors have likely also influenced spreads. One is the changing structure of corporate bond markets since the 2008 financial crisis. The post-crisis period has seen these markets undergo a significant transformation, adopting modalities that were once limited to other asset classes, mainly equity markets. This shift has two key, interrelated, aspects: the investor base and the trading structure. First, the investor base has broadened beyond traditional participants such as pension funds and insurance companies, with mutual funds and ETFs playing an increasingly important role. Relatedly, many aspects of bond trading in secondary markets have been reshaped. These changes are largely driven by the electronification of trading and extend to the role of dealers, the trading protocols used, and the market infrastructure required to support these new trading practices.

Bond markets have historically been slower than equity markets to adopt electronic trading. This partly reflects structural features that make automation more challenging, particularly in corporate bond markets. As opposed to equities, companies typically have numerous outstanding bonds that differ in maturity, coupon and credit characteristics, limiting the standardisation of the trading process. For instance, at the beginning of 2026, non-financial companies included in the S&P 500 index had a total of 443 outstanding equities, with only three firms having more than one class of shares. By comparison, the same companies had 7 459 outstanding bonds (Figure 2.12). The comparison is even starker when including financial companies – some of which have tens of thousands of bonds outstanding.

Figure 2.12. Number of outstanding shares and bonds issued by non-financial S&P 500 companies
Copy link to Figure 2.12. Number of outstanding shares and bonds issued by non-financial S&P 500 companies

There are more than 16 bonds for every non-financial equity in the S&P 500

The trading structure also differs between corporate and government bond markets. Whereas the latter, which serves as a benchmark for essentially all financial market pricing, adopted electronic and algorithmic inter-dealer trading in on-the‑run securities in the United States in the mid‑2000s, corporate bond markets continued to rely on more traditional models for significantly longer (US Senate, Committee on Banking, Housing, and Urban Affairs, 2016[4]). As a consequence, corporate bond trading activity has historically been fragmented across a large number of instruments, with only a small share trading frequently, resulting in persistently low liquidity. This illiquidity was reinforced by an investor base dominated by institutional investors such as pension funds and insurance companies, whose buy-and-hold strategies add very little to secondary-market turnover.

Banks historically played a very central role in the dealer-intermediated, over-the‑counter market model by warehousing bonds on their balance sheets and providing liquidity to investors when needed. This intermediation function was especially important given the high degree of heterogeneity across outstanding securities. However, the role of banks has changed significantly in recent years. In the aftermath of the global financial crisis, tighter banking regulations reduced banks’ incentives to hold large inventories of risky assets, including corporate bonds (see Chapter 3). As a result, they have become less central to market intermediation, creating space for non-bank financial institutions (NBFIs) to play a larger role (Figure 2.13). This is notably the case for investment funds and ETFs, which have expanded their footprint significantly (OECD, 2024[5]). Hedge funds have also increasingly begun trading securities on their own accounts, absorbing some of the risks previously managed by banks, while banks have shifted towards providing leverage to support these activities (Eren and Wooldridge, 2021[6]). This change in roles has contributed to the increase in hedge fund leverage observed in recent years.

Figure 2.13. . Trends in dealer positions, investment fund activity and hedge fund leverage
Copy link to Figure 2.13. . Trends in dealer positions, investment fund activity and hedge fund leverage

Funds have filled large part of the gap left when bank dealers withdrew from corporate bond warehousing

Because their incentives and structures are more conducive to frequent trading, this new investor base has helped shift bond market trading to more closely resemble the equity market model, notably by accelerating electronification. By November 2025, electronic trading accounted for around 50% of corporate bond trading in the US market (Barclays, 2025[7]). The prevalence of ETFs, notably, has enhanced the tradability of underlying bonds and lowered transaction costs by allowing intraday trading on electronic platforms. Their expansion has also been an important enabler of a move towards portfolio trading, as ETFs provide a price and liquidity reference for baskets of bonds. Market makers can therefore price and trade portfolios that resemble an ETF’s composition using the ETF as a benchmark. As a result, bonds with higher ETF ownership are more likely to be traded through portfolio trades, especially when they are relatively illiquid. This has had beneficial effects on liquidity. Research suggests that portfolio trading can reduce transaction costs by more than 40% on average compared to request for quote (RFQ) trading, with the largest gains observed for the least liquid bonds (Meli and Todorova, 2023[8]). This boost is most pronounced for less liquid bonds because portfolio trading allows investors to trade baskets of securities as a single risk exposure rather than as individual instruments, thereby facilitating transactions in bonds that would otherwise be difficult to trade.

This illustrates a dynamic of mutually reinforcing trends, where a new investor base is more prone to trade than the previous one, which ushers in changes to trading structures, further increasing those investors’ willingness to participate in the market. More recently, increased trading and liquidity has facilitated the entry of a new set of participants. Most notably, principal trading firms (PTFs) have emerged as liquidity providers in electronic bond markets. PTFs trade on their own account using high-frequency, automated strategies and compete primarily on execution speed. Unlike traditional dealers, they do not warehouse bonds, relying instead on rapid intraday turnover supported by short-term leverage from prime brokers and banks (Eren and Wooldridge, 2021[6]). This business model is best suited to deep and liquid markets, as it depends on the ability to offload inventory quickly. As a result, PTFs initially focussed on equity markets, where electronic trading and liquidity were already well established. Over time, they expanded into other asset classes, particularly standardised instruments such as on-the‑run US Treasuries. More recently, major PTFs have begun operating in corporate bond markets, reflecting both improvements in market liquidity and increasing sophistication in electronic trading strategies.

It should be noted that while dealers have become less central to bond warehousing, they remain key participants in corporate bond markets. Their role continues to be crucial even in electronic trading environments, as the dominant execution protocol, the RFQ model, relies heavily on dealer intermediation. In an RFQ protocol a customer requests price quotes for specific bonds from several dealers. Dealers submit quotes within a set time window, and the customer selects the best offer to execute the trade. An alternative protocol that has emerged in bond markets is all-to‑all trading, in which any market participant can trade directly with any other, without the need for dealer intermediation, but although it has grown noticeably in recent years it still represents a minor share of overall trading (The Trade, 2021[9]; 2025[10]).

Figure 2.14 provides a stylised overview of these developments, illustrating the change in key market participants and the modalities with which they trade.

Figure 2.14. Stylised overview of the evolution of corporate bond market dynamics
Copy link to Figure 2.14. Stylised overview of the evolution of corporate bond market dynamics

Corporate bond markets have evolved significantly since 2008, with new actors ushering in new trading structures

The move from bank-intermediated, single‑bond trading to an increasingly electronic and multifaceted structure with several intermediaries and the rise of portfolio trading has injected liquidity into what was traditionally an illiquid market dominated by buy-and-hold investors. Evidence suggests that the growing use of electronic trading, together with the reduced role of banks in trading intermediation, has improved the efficiency of corporate bond markets by lowering transaction costs and increasing price transparency (O’Hara and Alex Zhou, 2021[11]). Consistent with these developments, bid-ask spreads, a commonly used measure of market liquidity, have narrowed substantially over time, declining from an average (median) of 77 (81) basis points of par value at the beginning of 2013 to 33 (30) basis points at the end of 2025 (Figure 2.15). This trend holds when looking at a group of the same bonds over time and does therefore not appear to be driven by composition effects. These improvements have supported more active secondary-market trading. The share of outstanding investment grade corporate bonds in the United States that did not trade on a weekly basis dropped from 30% in 2016 to 10% in 2025 (Todorova, 2025[12]).

Figure 2.15. Evolution of bid-ask spreads
Copy link to Figure 2.15. Evolution of bid-ask spreads

Bid-ask spreads have reduced significantly over time for a large sample of corporate bonds

Illiquidity of corporate bonds compared to sovereign bonds has traditionally been a significant contributor to corporate spreads. Previous work in the context of the traditional, dealer-intermediated model (see e.g. Huang and Huang (2012[13]), Longstaff, Mithal and Neis (2005[14]), Chen, Lesmon and Wei (2007[15])) identifies a significant part of credit spreads that cannot be explained by credit risk alone, often attributing this to (il)liquidity premia. All else equal, then, the increase in liquidity illustrated above would seem to imply a compression in credit spreads through a decrease in the liquidity premium.

Other technical factors

Exogenous and differential variation in demand and supply is another key factor. Since 2008, corporate bond funds have seen strong net inflows, becoming one of the most significant investor categories in major corporate bond markets globally (OECD, 2024[5]). When these funds see inflows, their purchases of corporate bonds increase near-mechanically. However, while net issuance tapered off sharply in response to interest rates increases in 2022, there was no proportional decrease in the assets under management of the funds invested in these markets. That means funds need to reinvest maturing bonds into a smaller market with the consequence of putting downward pressure on credit spreads. That is particularly true for higher-risk bonds. While non-investment grade net issuance has been negative or near-zero in every year since 2022, net fund flows rebounded strongly in 2024 and remained significant in 2025 (Figure 2.16). Even at net zero fund inflows, negative net bond issuance (i.e. decreasing outstanding amounts) implies demand rising relative to supply, given funds’ reinvestment needs as their existing holdings mature, depressing spreads. Positive fund flows with negative net issuance exacerbates that dynamic.

Figure 2.16. Investment fund inflows versus net issuance, non-investment grade bonds
Copy link to Figure 2.16. Investment fund inflows versus net issuance, non-investment grade bonds

Net flows to non-investment grade corporate bond funds have outpaced net issuance since 2022

Another significant shift in demand dynamics comes from major central banks’ quantitative tightening (QT) programmes (see Chapter 1). The magnitude of these programmes begs the question of whether they will impact corporate spreads. Industry analysis addressing this question suggests that because quantitative easing (QE) did not impact riskier corporate spreads (i.e. they did not differentially increase demand compared to government bonds), the same should be true for QT (Moody's Ratings, 2024[16]). However, research also suggests there is an asymmetry between QE and QT for sovereign bonds, meaning relationships observed during QE might not hold in reverse during QT (Jiang and Sun, 2024[17]).

There are also less evident factors to consider, such as how different investors’ demand interacts with changes in monetary policy. Specifically, the strength of the traditional dynamic whereby tighter monetary policy leads to a “risk-off” sentiment in markets, which in turn drives riskier corporate spreads upward, depends in part on the market’s investor composition. Domanski, Shin and Sushko (2015[18]) demonstrate how insurance companies, facing a duration mismatch as long-term interest rates fall, respond by increasing their holdings of longer-term debt, pushing long-term rates down and exacerbating the very same mismatch they sought to address. Reversely, when interest rates increase, the duration gap reverses and they reap equity gains, increasing their risk appetite, with consequent rebalancing of portfolios towards riskier corporate bonds that can lead to compression of credit spreads, as shown by Li (2025[19]) in the context of US life insurance companies. In settings where these companies are significant investors, this can change the dynamic for the entire market; in the post-2008 period, corporate bond spreads for the riskiest bonds in the United States have moved inversely with long-term interest rates. More broadly, this underscores the importance of the composition of a market’s investor base and the interest of monitoring changes thereto (see Chapter 3).

Figure 2.17 offers a conceptual overview of some technical factors that affect corporate spreads, illustrating in which direction, and how they interact with the monetary policy environment. It seeks to emphasise two key points. First, there is an array of factors unrelated to fundamental corporate credit quality that affect corporate credit spreads, sometimes diminishing the measure’s power in explaining actual corporate credit risk. Second, these factors sometimes move in different directions, even for a given monetary policy stance, increasing the difficulty of interpreting what a movement in spreads implies.

Figure 2.17. Technical factors affecting corporate credit spreads
Copy link to Figure 2.17. Technical factors affecting corporate credit spreads

In addition to credit fundamentals, several technical factors drive corporate spreads in different directions

Decomposing corporate credit spreads

Understanding the extent to which recent spread movements are driven by fundamental credit quality improvements as opposed to technical factors is critical to assessing risk in corporate credit markets and the corporate sector more broadly. Spreads on their own are surface‑level indicators; identifying the individual drivers that influence them allows for a significantly more informed policy stance. For example, if a spread compression is driven entirely by technical factors, it gives no meaningful signals about reductions in corporate credit risk. If the liquidity premium is constant, decreases in bid-ask spreads or alternative indicators might mask reductions in other measures of liquidity, ultimately calling into question whether companies reap the benefits of the new trading structure that has emerged since 2008. To gauge what factors have been driving spread movements, the following section decomposes credit spreads based on a structural credit model. The analytical approach and data used are summarised in Box 2.1 and further in Annex 2.C.

Box 2.1. Data and methodology for credit spread decomposition

Copy link to Box 2.1. Data and methodology for credit spread decomposition

The decomposition analysis uses a structural credit model in the style of Merton (1974[20]). It calculates a GZ spread (matching the bond’s cashflows to a fitted risk-free zero-coupon curve) for a large set of bonds representative of a major global corporate investment grade bond index. As in Gilchrist and Zakrajšek (2012[21]) and Gilchrist et al. (2021[22]), the spread is decomposed to isolate compensation for firm-specific default risk, but the model is expanded to include a liquidity premium as well, reflecting the compensation for trading and holding less liquid securities. The residual spread that cannot be explained by firm-specific default risk or liquidity – the excess bond risk premium – reflects risk aversion, the returns investors require to bear credit risk beyond bond-specific expected default losses, and liquidity.

To isolate this residual component, in addition to issuer-level default risk and bond-level liquidity, the regressions control for bond-level heterogeneity. Default risk is inferred from market and balance‑sheet information at the issuer level, resulting in a forward-looking proxy. Each month, bond spreads are then related to this default-risk proxy, a liquidity indicator based on bid – ask quotations, and standard bond characteristics (such as maturity, coupon, call structure, duration and age). Because the spread is constructed as the difference between the bond’s yield-to-maturity and a cash-flow-matched risk-free yield-to-maturity, bond characteristics enter mechanically into the spread through cash-flow timing and discounting. These characteristics – defined as the structural spread – are therefore included as controls to account for contractual and pricing differences across securities and to prevent the residual from capturing predictable cross-sectional variation rather than changes in risk pricing. The fitted component yields the part of spreads attributable to expected default risk, liquidity conditions and observable bond features, while the remaining component captures economy-wide shifts in the price of bearing credit risk. Taking the market value‑weighted monthly average of that residual component yields the excess bond premium, interpreted as a measure of credit market tightness and risk-bearing capacity beyond changes in expected defaults. More detailed explanations are available in Annex 2.C.

Data

The Bloomberg Global Aggregate Corporate Index is used as the reference index. Historical index constituents are proxied using the holdings of a major ETF tracking the index, launched in 2012. As shown in the first panel of Figure 2.18, the ETF’s price performance very closely tracks that of the reference index over time. Using the ETF’s historical constituents, retrieved from Morningstar and comprising over 25 000 corporate bonds, bond-level indicators (bid-ask quotes, coupons, maturities, call structures, etc.) are downloaded from LSEG. As shown in Panel B, the median monthly spread in the final bond-level dataset closely tracks the index-level option-adjusted spread. The sample is thus taken to be representative of the reference index. Bond-level identifiers are then mapped to equity identifiers for the ultimate parents of the issuing companies. These are used to download monthly equity prices (also from LSEG) of the ultimate parents, which are used to calculate the equity risk premium, as well as for other analyses (see Section “Are debt markets becoming more like equity markets?”). Following cleaning and exclusions, the credit spread decomposition analysis includes 6 600 USD-denominated bonds.

Figure 2.19 plots the results of the decomposition. It splits the results into a structural (capturing bond characteristics such as age and duration) and non-structural components (liquidity, expected default and excess bond risk). The latter is the component of interest, as it is the one driving changes in spreads over time (Panel A). Three developments in the non-structural component stand out (Panel B). First, the liquidity premium has been decreasing steadily over time. In absolute terms, it has been a major contributor to spread compression in recent years – of the 97‑basis point reduction in total spreads since 2013, 34 are attributable to reductions in liquidity premia. This is consistent with improvements in liquidity as measured by lower bid-ask spreads and changes in trading outlined above. It should be noted, however, that the liquidity proxy – bid – ask spreads – may co-move with credit spreads, particularly during stress episodes, which can give rise to simultaneity problems. There are also possible endogeneity issues, such as joint determination of spreads and liquidity where higher spreads can themselves give rise to greater illiquidity, or omitted variables that are captured in the liquidity variable, biasing the estimate upwards. The analysis uses bid-ask spreads as the observable proxy for trading frictions to improve the accounting decomposition of credit spreads – the results should therefore not necessarily be read as a causal effect of liquidity on spreads.

Second, the default risk component has actually increased somewhat over the last decade in basis point terms – in other words, the compensation required by investors for expected loss risk has increased. Thirdly, the excess bond risk premium – a measure of general risk aversion – has been negative since 2023, indicating a period of significant risk willingness among investors, driving spreads down to levels that cannot be explained by fundamentals, liquidity or bond characteristics.

Figure 2.19. Decomposition of corporate credit spreads
Copy link to Figure 2.19. Decomposition of corporate credit spreads

Reductions in liquidity premia and investor risk aversion have been major contributors to reductions in credit spreads

It is important to distinguish between changes in different components in absolute (basis point) terms and their share of the total spread. Figure 2.20 plots the share of each of the three components in total spreads over time. As shown in Panel A, while the liquidity premium has reduced significantly in absolute terms, its reduction as a share of total spreads has been less pronounced. The structural trend is still downward pointing, although with significant variation over time. This is consistent with other research. It bears noting that the dynamics may differ for investment grade (shown here) and non-investment grade bonds. Wu (2020[23]) points to an increase in the share of spreads made up by liquidity premia for non-investment grade bonds since 2008, manifesting not as bid-ask spreads, but as longer trading delays. In other words, the post-2008 reduction in bid-ask spreads for non-investment grade bonds does not necessarily indicate an increase in their liquidity, but could at least partially reflect a change in how illiquidity manifests. Corporate spreads have become more sensitive to variations in bid-ask spreads. The decrease in dealer intermediation and warehousing that followed from the introduction of the post-crisis Basel regulations have meant trades take longer to execute, which is a form of illiquidity.

Compensation for expected default losses now explain the greatest part of spreads (Panel B). This does not mean that default risk has increased substantially (as shown in Panel B of Figure 2.19, in absolute terms default risk compensation has remained at similar levels since 2019, although with spikes during the COVID‑19 pandemic and 2022 inflation shock) – it means that because of improvements in liquidity and greater risk-willingness among investors, corporate spreads now to a larger extent reflect default risk as opposed to other factors. The recent decrease in the excess bond risk premium is remarkable (Panel C). This residual, again, can be thought of as the general risk-willingness of investors. It spiked in 2020 during the “dash for cash” period at the onset of the pandemic, dropping sharply and contributing negatively to spreads during 2021 when extremely accommodative monetary policy led to a risk-on sentiment across asset classes, and finally spiking again during the inflation shock in 2022. Since then, it has dropped to the lowest level on record, potentially reflecting optimistic investor sentiment, in line with the buoyant prices and steep valuations seen in equity markets.

Figure 2.20. Spread components as a share of total spread
Copy link to Figure 2.20. Spread components as a share of total spread

As illiquidity and general risk premia have decreased, default risk now accounts for the greatest share of spreads

This decomposition helps interpret the recent extremely low levels of corporate credit spreads. Spread compression does not seem to be indicative of improved credit fundamentals measured as default risk. Instead, changes in trading structures and the participation of more trading-prone investors have increased liquidity, driving down the premium traditionally required by investors for holding less liquid securities rather substantially. Simultaneously, investor risk willingness has increased, putting further downward pressure on spreads. It is difficult to identify precisely what is driving this sentiment – developments in equity markets may suggest that it is because of an expectation that the macroeconomic effects of advancement in AI will be very positive, but movements in other assets (notably safe‑haven assets such as gold) can be interpreted as giving contradicting signals. Whatever the cause, the result is that corporate spreads increasingly reflect two factors: default risk and investor risk-willingness.

A final point should be emphasised with respect to liquidity. As outlined above, it seems plausible that decreased liquidity premia are related to changes in trading structures. It should therefore be noted that while the shift towards a more transactional and less relationship-based market structure has increased liquidity and improved market functioning in normal times, it may also weaken the market’s capacity to absorb large selling pressures in times of crisis.

Traditionally, banks were often willing to expand their balance sheets to accommodate client trades, partly to strengthen relationships with clients and securing future business (O’Hara and Zhou, 2025[24]). In contrast, with electronic trading becoming more prevalent, repeat-client relationships are less central, reducing the incentives of non-bank dealers, such as hedge funds, to hold corporate bonds during periods of stress. The deterioration in market liquidity during COVID‑19 in March 2020 highlights this shift (see the increase in the liquidity premium in 2020 in Figure 2.19). Selling pressures across fixed-income securities, including corporate bonds, led to sharp fund outflows and a substantial increase in transaction costs. Instead of supplying liquidity, dealers reduced inventories and became net sellers, further amplifying market stress. Alternative liquidity sources, such as electronic trading without dealer intermediation, became more important but remained limited and costly.

Ultimately, market stability depended on central bank intervention, which provided funding to dealers and directly purchased corporate bonds to restore liquidity (O’Hara and Zhou, 2021[25]). While market conditions, shock characteristics, and policy responses shape the outcomes of each episode of stress, it is still unclear whether technology-driven market structures providing greater liquidity in normal times come at the cost of heightened vulnerability during periods of market disruption.

The role of debt markets in financing the AI expansion

Copy link to The role of debt markets in financing the AI expansion

The continued development of increasingly sophisticated artificial intelligence (AI) systems will require enormous amounts of investment. Large parts of that investment will come from the technology sector, in particular from a relatively small number of frontier companies which have historically financed investment through internally generated cash flows and, to a lesser extent, equity issuance, with limited reliance on debt. Given the magnitude of the costs associated with AI expansion however, this funding model is increasingly misaligned with future investment needs. Consequently, debt markets will play a much more central role in funding technology companies than they previously have. This section illustrates that change and addresses its implications.

For much of the past two decades, the technology sector at large has been characterised by relatively low capital intensity, with spending concentrated on software development and research and development activities to a greater extent than other industries. In 2024, the median capital expenditure to sales ratio of technology firms globally stood at 2.2%, the lowest among all industries, as it has been consistently since 2003 (the ratio was higher in the early 2000s, reflecting exceptionally high infrastructure investment during the initial buildout of the industry at the turn of the century, including telecommunications networks, data centres and hardware capacity).

This asset-light model, coupled with strong and persistent cash flow generation, allowed technology firms to fund expansion without significant reliance on external borrowing. The ratio of free cash flow to capital expenditures, a measure of a firm’s ability to fund investment without external financing, has consistently been among the highest across industries (Figure 2.21).

Figure 2.21. Internal financing capacity by industry
Copy link to Figure 2.21. Internal financing capacity by industry

Historically, tech firms have been characterised by relatively low physical capital intensity and strong free cash flows

A group of nine firms – commonly called hyperscalers – primarily US technology companies, are at the core of the AI investment boom (Figure 2.22). Their funding models have historically been even more geared towards internal funds than the rest of the already relatively capital-light technology sector, with free cash flow to capital expenditure ratios well above the industry average. However, 2025 saw a shift in this paradigm, with the majority of hyperscalers increasing capex-to-sales ratios substantially above historical averages. In some cases, capital expenditure as a share of revenue increased over tenfold compared to historical averages, pushing capex intensity above that of the aggregate of the traditionally very asset-heavy utilities sector. There have been corresponding reductions in free cash flow ratios, and in some cases increases in leverage. Given the historically low levels of debt funding among these firms, leverage still broadly remains manageable, but the increasing trend is clear.

Figure 2.22. Hyperscalers’ cash flow capacity and leverage
Copy link to Figure 2.22. Hyperscalers’ cash flow capacity and leverage

Forefront AI companies are no longer financing their capital expenditure with internally generated cash flows alone

The change in hyperscalers’ funding model is clearly visible in corporate bond markets. In 2025, they issued a total of USD 122 billion (of which USD 88 billion over 54 days in the last few months of the year), equivalent to 45% of total issuance by all technology firms globally, the largest amount on record in real terms and over 3 times more than the historical annual average since 2000 (Figure 2.23, Panels A and B). Some firms issued multiples of their historical averages in 2025 (Panel C). The figures likely understate the true scale of hyperscaler-related financing activity as they capture only bonds issued directly by the companies themselves. However, firms increasingly turn to special purpose vehicles (SPVs) for bond issuances, allowing them to keep the debt separate from their main balance sheet (Moody's, 2026[26]). The rise of such financial structures to secure AI investment capital is exemplified by Meta’s USD 27 billion SPV debt deal with private capital firm Blue Owl Capital in October 2025, examined in greater detail in Box 2.2.

Figure 2.23. Corporate bond issuance by major hyperscalers
Copy link to Figure 2.23. Corporate bond issuance by major hyperscalers

Hyperscalers’ bond market borrowing has reached record volumes, driven by large‑scale AI investment needs

While these bond issuances represent significant amounts for single firms, the cumulative issuance in 2025 of USD 122 billion represents no more than 15% of total investment grade issuance by non-financial US firms and has therefore been absorbed without market-wide friction (although credit risk metrics of individual firms have sometimes responded sharply). However, hyperscalers’ issuance in 2025 represents only a fraction of projected future capex, which is extraordinarily large in some cases. Between 2026-2030, consensus estimates are for cumulative capital expenditure of USD 4.1 trillion (Figure 2.24, Panel A). The four largest players alone are expected to spend 3.5 trillion. To put that into perspective, annual capital expenditure by all non-financial corporations in the United States in 2025 (annualised at the third quarter) was just above USD 3 trillion (US Federal Reserve, 2026[27]).

Large parts of this will inevitably need to be financed through corporate debt markets, as the magnitude exceeds what even the largest technology firms can reasonably finance through internal cash generation and equity alone. Recent industry estimates identify a financing gap as high as USD 1.5 trillion between 2025 and 2028, within an estimated USD 2.9 trillion in total capital expenditure requirements for AI-related infrastructure. Nearly four‑fifths (77%) of the required external funding is expected to be financed through debt markets (Morgan Stanley, 2025[28]). To gauge the impact on corporate bond markets, Panel B of Figure 2.24 plots the hyperscalers’ future capex-related bond funding as a share of global average annual non-financial firm gross issuance from 2020-2025 for different assumptions about the share financed through bond markets. In the conservative base case, where 29% of capex is funded through bond markets (equivalent to the average bond issuance as a share of capex from 2020‑2025), these nine firms alone – out of 9 235 firms issuing bonds globally in 2025 – would issue the equivalent to an average of 9% of historical global gross issuance from 2026-2030. If half of capex was to be financed through bond markets, in 2030 it would be equivalent to 15% of historical gross issuance. It bears noting that the historical average used includes 2020 and 2021, both record years for corporate debt issuance.

Figure 2.24. Hyperscalers’ capital needs and bond market capacity
Copy link to Figure 2.24. Hyperscalers’ capital needs and bond market capacity

Hyperscalers’ estimated capex needs would likely require substantial bond market borrowing

Bond markets reflect only one part of the debt financing used to fund the AI expansion. It spans several other markets as well; for data centre construction, for example, banks typically provide construction financing, primarily through syndicated loans. Private credit frequently supplements this financing, enabling developers to leverage their equity further in the construction phase. Upon completion, owners often refinance the project through single‑asset single‑borrower ABS or private placement bonds (Goldman Sachs, 2025[29]).

Private credit is set to become a critical funding source for AI expansion more broadly, building on momentum already visible in 2025. AI-related private credit transactions reached USD 59 billion in 2025, a near sevenfold increase from 2024, and a USD 17 billion increase from the second-highest year on record, 2022 (Figure 2.25, Panel A). The surge was driven by higher deal values rather than more deals. The growth in AI private credit activity reflects both broader market expansion – total private credit deal value increased from USD 97 billion to USD 174 billion – and greater allocation to AI, with the sector’s share of deals rising from 9% in 2024 to 34% in 2025 (Panel B). The role played by private credit markets in financing the AI expansion, and their increasing interconnection with traditional bond markets, are further examined in Box 2.2.

Figure 2.25. Private credit deals involving AI companies
Copy link to Figure 2.25. Private credit deals involving AI companies

The aggregate size of private credit deals involving AI companies in 2025 was the highest on record

Box 2.2. Private credit, the AI expansion and the blurred lines between different debt markets

Copy link to Box 2.2. Private credit, the AI expansion and the blurred lines between different debt markets

Private credit, an asset class that totalled no more than USD 1 trillion globally as recently as 2018, is expected to supply USD 800 billion to the AI expansion alone over the next four years, primarily through asset-based finance (ABF) structures (Morgan Stanley, 2025[28]; OECD, 2025[2]). Private credit markets have proven to offer issuers the opportunity to structure more complex deals tailored to specific needs than traditional bond markets. Over time, they have grown to accommodate larger and larger volumes. However, the size of AI-related investment needs and associated debt financing are too significant to be financed by private credit markets alone. This has led borrowers to combine private credit and traditional bond markets, increasingly blurring the distinction between the two.

A recent transaction exemplifies this. In October 2025, Meta announced the creation of a joint venture with funds affiliated with Blue Owl, a private capital firm, to develop a major data centre campus. The total committed funding for the project amounts to roughly USD 29 billion. Of this amount, funds affiliated with Blue Owl Capital provided USD 23 billion in exchange for 80% of the interest in the joint venture, while Iris Crossing LLC, a wholly owned subsidiary of Meta, provided the remaining USD 6 billion, retaining 20% ownership (Figure 2.26). A wholly owned subsidiary of the joint venture, Laidley LLC, will own and operate the data centre campus. The capital committed by Blue Owl Capital consists primarily of proceeds from a USD 27 billion bond issuance by Beignet Investor LLC, a bankruptcy remote vehicle wholly owned by a subsidiary of Blue Owl Capital. The privately placed issue will be amortised with repayments commencing upon completion of the campus in 2029 and extending through 2049.

A notable feature of this arrangement is the residual value guarantees provided by Meta, compensation for potential shortfalls in the market value of the data centre versus a contractually agreed threshold in case of non-renewal or termination of the lease. These can reach up to USD 28 billion but are not recorded on its balance sheet. This deal structure illustrates how the AI expansion’s significant investment needs may lead to a less clear demarcation between market segments and possibly an increase in the complexity of corporate debt deals.

Figure 2.26. Organisational structure of joint venture between Meta-Blue Owl Capital
Copy link to Figure 2.26. Organisational structure of joint venture between Meta-Blue Owl Capital

In addition to spanning different debt market segments, AI-related investment needs stretch far beyond the hyperscalers and other technology companies that own and run major AI models. Scaling AI also increases financing needs further up the supply chain, notably for energy supply. A significant part relates to the construction and operation of high-capacity data centres, the critical infrastructure for the AI expansion. Much of the existing stock of data centres was built before the rapid rise of artificial intelligence and are increasingly ill-fitting for energy and computing intensive AI workloads, requiring high-powered graphic processing units (GPUs). At present, most operational data centres are optimised for cloud computing: cloud workloads make up 58% of current demand and AI workloads 13%. This balance is projected to shift significantly, with AI demand rising by a further 15 p.p. by 2027, largely displacing both cloud and traditional workloads (Goldman Sachs, 2025[31]). Meeting this structural change will require substantial investment in new, AI-ready facilities.

Recent industry estimates suggest that meeting global demand for AI-related computing power will require approximately USD 5.2 trillion in investment worldwide by 2030 (Figure 2.27). It is critical to emphasise that these estimates do not include investment needs by cloud providers (such as the hyperscalers) or AI developers. The bulk of this is expected to come from technology developers and designers, such as semiconductor manufacturers and IT hardware suppliers that produce the chips and computing equipment used in data centres. These firms are projected to spend approximately USD 3.1 trillion in AI-related capital expenditures, covering GPUs, central processing units (CPUs), memory system services and rack hardware. Companies involved in energy supply (e.g. energy providers, power generators, cooling and electrical equipment manufacturers and telecommunications operators) are forecasted to invest USD 1.3 trillion. Lastly, USD 800 billion is expected for data centre construction, encompassing land acquisition, materials, and skilled labour (McKinsey & Company, 2025[32]). In other words, in addition to the substantial expected use of debt markets by hyperscalers, the AI expansion will likely also see significantly increased bond issuance across industries spanning real estate, energy and hardware development. This calls into question the ability of the currently USD 17.2 trillion global non-financial corporate bond market to absorb new supply of this magnitude, especially in a context of still-expanding sovereign bond borrowing and a changing investor base (see Chapters 1 and 3).

Figure 2.27. Projected data centre investment (excl. cloud providers and AI developers), 2025‑2030
Copy link to Figure 2.27. Projected data centre investment (excl. cloud providers and AI developers), 2025‑2030

Data centre investments are projected to reach USD 6.7 trillion by 2030, of which USD 5.2 trillion for AI

Are debt markets becoming more like equity markets?

Copy link to Are debt markets becoming more like equity markets?

The developments identified in this chapter raise a more general question: are corporate debt markets becoming more equity-like? This section looks at two aspects in which that might be happening – issuer-level concentration and market pricing – and the implications thereof.

Issuer-level concentration and risk characteristics

The expected increase in debt issuance by a small number of firms associated with the AI expansion would likely move corporate debt markets towards greater concentration. In contrast to public equity markets, where this trend has long been visible – ten US companies now make up a quarter of the broadly tracked MSCI World Index – corporate bond markets have generally not seen corresponding increases in single‑company exposure (partly a natural consequence of the payoff structure of bonds and equities). Globally, about a third of total outstanding debt is owed by the 100 largest issuers, a figure that has remained largely constant for the past 25 years (OECD, forthcoming[33]; 2025[2]). However, given projected capital expenditures by individual firms of the magnitudes shown in Figure 2.24, that is liable to change, particularly when considering that these firms currently typically operate at much lower leverage than the corporate sector at large, giving them greater ability to sustain increased debt issuance.

Notably, the firms that are expected to represent growing shares of corporate bond markets are to a large extent the same ones that are already dominating public equity markets. At the end of 2025, nine major hyperscalers represented 12% of global equity market capitalisation and 24% of US market capitalisation, exceeding the entire Chinese market and approaching that of Europe (Figure 2.28).

Figure 2.28. Hyperscalers’ share in global equity market capitalisation
Copy link to Figure 2.28. Hyperscalers’ share in global equity market capitalisation

Hyperscalers accounted for 12% of global market capitalisation at the end of 2025

The projected increases in borrowing by hyperscalers will likely provide further momentum to an ongoing trend where the technology sector makes up an increasingly large share of the total non-financial bond market. In 2025, the technology sector’s share of global issuance reached 9.6%, the highest level since 2000 and a 3.9 p.p. increase from 2024 (Figure 2.29, Panel A). The technology sector is already one of the most heavily concentrated sectors when it comes to outstanding bond debt, second only to the telecom sector, as measured using a Herfindahl-Hirschman index (Panel B). Crucially, the technology sector represents a ten times larger share of total outstanding bond debt than telecom, ranking third overall behind only industrials and utilities – both significantly less concentrated sectors (Panel C). Current developments are therefore aligned with increasing corporate bond market concentration.

Figure 2.29. Concentration in the non-financial corporate bond market
Copy link to Figure 2.29. Concentration in the non-financial corporate bond market

The technology sector is relatively concentrated and makes up an increasing share of total non-financial issuance

That, in turn, exposes index-based investors to greater risk of single‑name defaults. Because the hyperscalers are predominantly highly profitable, highly rated companies, this risk appears minor at present, but the tendency towards increasing exposure concentration in corporate bond markets is nevertheless real. Credit risk, as measured by credit default swap (CDS) spreads, have recently increased for some hyperscalers. However, it is not possible to distinguish the extent to which this refers to concerns about the creditworthiness of the issuers in question as opposed to an increased willingness to buy insurance as concentration risk has increased. In addition, the single‑name CDS market is relatively illiquid, and spreads are sensitive to trading activity, so these developments cannot straightforwardly be interpreted as pure increases in market perceptions of credit risk (IOSCO, 2025[34]).

A more pressing question relates to whether the nature of credit risk is changing for an important part of the debt markets. The substantial capital deployed for investment in AI technology and infrastructure rests on projections of future profits, which in turn hinge on widespread AI implementation. These projections, while plausible, may reflect an equity-like risk which major companies are partly financing with significant fixed debt obligations. This may be exacerbated by the changing composition of the underlying asset serving as effective collateral for much of the borrowing – data centres. In the cloud computing era, investment costs were largely concentrated in the physical building shell and mechanical, electrical and plumbing systems, enabling financial arrangements similar to traditional real estate, with tangible assets serving as collateral. By contrast, AI data centres invert this cost profile: high-performance compute hardware typically represents three to four times the value of the physical facility, while rapid technology advances introduce uncertainty around the long-term value of these assets (Goldman Sachs, 2025[29]). This shift complicates collateralisation and risk assessment for lenders, changing the effective collateral tied to the debt from the physical data centre itself to the cash flow generated by companies that contractually commit to leasing the facility upon completion. This, again, effectively amounts to a more equity-like risk structure, but with the nominal payment requirements of a debt instrument.

Market pricing

Another aspect in which debt markets might be becoming more equity-like has to do with pricing. Index-based hedge ratios – measuring the sensitivity (conditional co-movement) of corporate credit spreads to changes in equity prices – appear to have become increasingly negative for broad-based indices over time (Figure 2.30). That means that when index equity prices increase (decrease), index corporate credit spreads compress (widen) more. There is significant variation over time, and as is clearly visible in the figure, this sensitivity spikes in times of financial turmoil (see 2008 and 2020), consistent with a “flight to safety” dynamic where portfolios rebalance towards safe assets such as government securities (as implied by corporate credit spreads widening, see Figure 2.1). Even so, the downward drift appears more structural, holding over long periods of time, begging the question of what is driving this dynamic.

Figure 2.30. Investment grade credit spread sensitivity to changes in equity prices (hedge ratios)
Copy link to Figure 2.30. Investment grade credit spread sensitivity to changes in equity prices (hedge ratios)

Spreads have increasingly co-moved with equity prices recently, compressing more in response to higher prices

Two possible explanations of this development merit consideration in light of the trends presented in this chapter. First, as outlined above, bond markets have seen a concurrent shift in trading practices and the investor base, especially since the 2008 financial crisis (see also Chapter 3). The participation of more trading-prone investors and advances in trading modalities have reinforced each other, with the result of greater trading activity in corporate bond markets. It is possible that this has led to greater homogeneity in trading patterns between equities and corporate bonds in a more general sense, increasing asset co-movement at least in the short and medium term. If that is the case, the benefits of increased liquidity and improvements in price finding stemming from these changes must be weighed against the possibility of higher correlations which might exacerbate market stress, in a worst-case scenario leading to broad-based simultaneous unwinding of positions, with the prospect of propagation to the real economy.

Figure 2.31 plots correlations of price returns (as opposed to spreads) between corporate bonds and equities for two different sets of companies: firstly, for bonds and equities of the same companies, based on the index proxy described in Box 2.1 (Panel A) and secondly, for the same broad indices shown in Figure 2.30 (Panel B). Both sets of companies show increasing correlations between the 2010s and the post-2022 period, consistent with increasing co-movement as implied by the hedge ratios in Figure 2.30. As would be expected, the price sensitivity between assets is higher for the same‑company sample. Albeit with different seniority and rights attached, a share and a bond issued by the same company are ultimately claims on the same underlying assets, so it is reasonable for prices to move together, at least in standard settings where bond and equity investors are not in direct conflict, such as when a majority shareholder acts in ways that are counter to creditor interests in a context of weak investor protection. This is in line with previous research (e.g. Liu and Clarkson (2025[35]) and Kwan (1996[36])). However, there is no particular reason to expect fundamental valuations of same‑company prices to become more strongly correlated over time – the fact that this has happened might therefore reflect more frequent trading and better liquidity in bond markets. Notably, the post-2022 period is one in which portfolio trading has seen significant growth (Barclays, 2024[37]; MarketAxess, 2025[38]).

It is not necessarily expected that increased trading and liquidity should have differential effects on the two different samples, however. Yet the increase in sensitivity (from lower levels) for broad-based index returns is more than twice as high as that for sample‑company assets (Panel C).

Figure 2.31. Co-movements in corporate bond and equity prices
Copy link to Figure 2.31. Co-movements in corporate bond and equity prices

Broad-based indices of corporate bond and equity prices show stronger positive correlations since 2022

This might be related to a second possible driver. If the constituents of global bond and equity indices have converged – that is to say if investing in these indices increasingly means having claims on the same set of companies – whether due to increasing concentration or some other factor, that would help explain the differential change in correlation. Given the secular increase in passive investment, that would mean the different parts of global index investors’ portfolios have become increasingly correlated, diminishing their power as a hedge and posing the same type of self-reinforcing sales pressures in a downturn as discussed above.

It should be emphasised that both these explanations are based on the trends explored throughout this chapter; the data in Figure 2.31 alone are not enough to draw any conclusions about the actual underlying drivers, and the difference in the change in sensitivity is not necessarily statistically robust. There are numerous other plausible explanations that could provide similar outcomes; the literature studying the changing price correlation between bonds and equities is rich. Most such studies take a portfolio management perspective and therefore typically compare co-movements between corporate equity prices and safe‑asset government, rather than corporate, bonds. The two have been strongly negatively correlated pricewise for most of the 21st century, although with significant short-term variation, influencing global portfolio decisions. The traditional government bond-equity price correlation has recently gone from negative for most of the 2000s to positive. This holds across different markets, particularly in the post-2022 period. That new correlation runs counter to the hedging logic of many traditional bond-equity portfolios, but from a longer-term perspective, it amounts to a reversal to historical trends. The bond-equity price correlation only turned negative around 2000, before which it was primarily positive for nearly a century (Roncalli, 2025[39]; Rankin and Shah Idil, 2014[40]). One identified driver of this is changes in the macroeconomic setting and monetary policy conduct (Campbell, Pflueger and Viceira, 2014[41]). It is possible that this is what explains the developments in corporate bond-equity pricing co-movements shown in this section too. Nevertheless, as narrative explanations, changes in trading structure and increases in concentration are aligned with the data presented in previous sections and merit further study and consideration by policymakers, given the possibility of adverse effects on market functioning.

References

[7] Barclays (2025), The equitification of credit - Illiquid no more, FICC: Credit & Macro Research.

[37] Barclays (2024), Portfolio trading - Every seven minutes, FICC Research: Credit Strategy.

[3] BIS (2025), Annual Economic Report, https://www.bis.org/publ/arpdf/ar2025e.pdf.

[41] Campbell, J., C. Pflueger and L. Viceira (2014), Macroeconomic Drivers of Bond and Equity Risks, National Bureau of Economic Research, Cambridge, MA, https://doi.org/10.3386/w20070.

[15] CHEN, L., D. LESMOND and J. WEI (2007), “Corporate Yield Spreads and Bond Liquidity”, The Journal of Finance, Vol. 62/1, pp. 119-149, https://doi.org/10.1111/j.1540-6261.2007.01203.x.

[1] Davis, S. (2016), An Index of Global Economic Policy Uncertainty, NBER Working Paper Series, http://www.nber.org/papers/w22740.

[18] Domanski, D., H. Shin and V. Sushko (2015), “The hunt for duration: not waving but drowning?”, BIS Working Papers No. 519, https://www.bis.org/publ/work519.htm.

[6] Eren, E. and P. Wooldridge (2021), Non-banking financial institutions and the functioning of government bong markets, BIS Papers, https://www.bis.org/publ/bppdf/bispap119.pdf?utm_.

[22] Gilchrist, S. et al. (2021), “The Term Structure of the Excess Bond Premium: Measures and Implications”, Federal Reserve Bank of Atlanta, Policy Hub, Vol. 2021/12, https://doi.org/10.29338/ph2021-12.

[21] Gilchrist, S. and E. Zakrajšek (2012), “Credit Spreads and Business Cycle Fluctuations”, American Economic Review, Vol. 102/4, pp. 1692-1720, https://doi.org/10.1257/aer.102.4.1692.

[31] Goldman Sachs (2025), How AI is transforming data centers and ramping up power demand, https://www.goldmansachs.com/insights/articles/how-ai-is-transforming-data-centers-and-ramping-up-power-demand.

[29] Goldman Sachs (2025), Powering the AI era, https://www.goldmansachs.com/what-we-do/investment-banking/insights/articles/powering-the-ai-era/report.pdf.

[13] Huang, J. and M. Huang (2012), “How Much of the Corporate-Treasury Yield Spread Is Due to Credit Risk?”, Review of Asset Pricing Studies, Vol. 2/2, pp. 153-202, https://doi.org/10.1093/rapstu/ras011.

[34] IOSCO (2025), Single-Name Credit Default Swaps Market, https://www.iosco.org/library/pubdocs/pdf/IOSCOPD806.pdf.

[17] Jiang, Z. and J. Sun (2024), “Quantitative Tightening with Slow-Moving Capital”, NBER Working Paper No. 32757, https://doi.org/10.2139/ssrn.4558016.

[36] Kwan, S. (1996), “Firm-specific information and the correlation between individual stocks and bonds”, Journal of Financial Economics, Vol. 40/1, pp. 63-80, https://doi.org/10.1016/0304-405x(95)00836-4.

[35] Liu, L. and P. Clarkson (2025), “Asset Allocation, Diversification, and Co-Movement Effects: A Global Analysis of Bonds and Equities Issued by the Same Firm”, Journal of Asset Management, Vol. 26/6, pp. 615-641, https://doi.org/10.1057/s41260-025-00422-2.

[19] Li, Z. (2025), Long Rates, Life Insurers, and Credit Spreads, Working Paper, https://ziang-li.github.io/files/Bond_Credit_Spreads_Paper.pdf.

[14] LONGSTAFF, F., S. MITHAL and E. NEIS (2005), “Corporate Yield Spreads: Default Risk or Liquidity? New Evidence from the Credit Default Swap Market”, The Journal of Finance, Vol. 60/5, pp. 2213-2253, https://doi.org/10.1111/j.1540-6261.2005.00797.x.

[38] MarketAxess (2025), AxessPoint: Portfolio Trading and price discovery in US IG and HY markets, https://www.marketaxess.com/article/axesspoint-portfolio-trading-and-price-discovery-in-us-ig-and-hy-markets.

[32] McKinsey & Company (2025), The cost of compute power: a $7 trillion race to scale data centers, https://www.mckinsey.com/industries/technology-media-and-telecommunications/our-insights/the-cost-of-compute-a-7-trillion-dollar-race-to-scale-data-centers.

[8] Meli, J. and Z. Todorova (2023), Portfolio Trading in Corporate Bond Markets, https://papers.ssrn.com/sol3/papers.cfm?abstract_id=4295453.

[20] Merton, R. (1974), “ON THE PRICING OF CORPORATE DEBT: THE RISK STRUCTURE OF INTEREST RATES*”, The Journal of Finance, Vol. 29/2, pp. 449-470, https://doi.org/10.1111/j.1540-6261.1974.tb03058.x.

[26] Moody’s (2026), Hyperscalers’ reported AI-related lease commitments may understate economic risk.

[16] Moody’s Ratings (2024), Quantitative tightening may have little impact on credit spreads.

[28] Morgan Stanley (2025), How credit markets could finance AI’s trillion dollar gap, https://www.morganstanley.com/insights/podcasts/thoughts-on-the-market/credit-markets-ai-financing-gap-vishy-tirupattur-vishwas-patkar.

[24] O’Hara, M. and X. Zhou (2025), “US Corporate Bond Markets: Bigger and (Maybe) Better?”, Journal of Economic Perspectives, Vol. 39/2, pp. 215-234, https://doi.org/10.1257/jep.20251439.

[2] OECD (2025), Global Debt Report 2025: Financing Growth in a Challenging Debt Market Environment, OECD Publishing, Paris, https://doi.org/10.1787/8ee42b13-en.

[5] OECD (2024), Global Debt Report 2024: Bond Markets in a High-Debt Environment, OECD Publishing, Paris, https://doi.org/10.1787/91844ea2-en.

[33] OECD (forthcoming), Capital Markets and Competitiveness, OECD Directorate for Financial and Enterprise Affairs Working Papers.

[11] O’Hara, M. and X. Alex Zhou (2021), “The electronic evolution of corporate bond dealers”, Journal of Financial Economics, Vol. 140/2, pp. 368-390, https://doi.org/10.1016/j.jfineco.2021.01.001.

[25] O’Hara, M. and X. Zhou (2021), “Anatomy of a liquidity crisis: Corporate bonds in the COVID-19 crisis”, Journal of Financial Economics, Vol. 142/1, pp. 46-68, https://doi.org/10.1016/j.jfineco.2021.05.052.

[40] Rankin, E. and M. Shah Idil (2014), A Century of Stock-Bond Correlations, Reserve Bank of Australia, https://www.rba.gov.au/publications/bulletin/2014/sep/pdf/bu-0914-8.pdf.

[39] Roncalli, T. (2025), Rethinking the Stock-Bond Correlation, Amundi Investment Institute, https://research-center.amundi.com/files/nuxeo/dl/eb84a66b-ebac-4901-8d70-70f3ed02a0d4?inline=.

[30] S&P (2025), Beignet Investor LLC’s $27.3 billion senior secured debt assigned preliminary “A+” rating; Outlook stable, https://www.spglobal.com/ratings/en/regulatory/article/-/view/sourceId/101651795 (accessed on 16 January 2026).

[10] The Trade (2025), All-to-all trading thrives as markets face volatility, report finds, https://www.thetradenews.com/all-to-all-trading-thrives-as-markets-face-volatility-report-finds/.

[9] The Trade (2021), All-to-all corporate bond trading on the rise, report finds, https://www.thetradenews.com/all-to-all-corporate-bond-trading-on-the-rise-report-finds/?utm_.

[12] Todorova, Z. (2025), Why the bond market still relies on voice, Financial Times, https://www.ft.com/content/287f5f16-2224-4bd0-9e6e-d965ca2cee5b.

[42] US Department of Justice - Antitrust Division (2024), Herfindahl-Hirschman Index, https://www.justice.gov/atr/herfindahl-hirschman-index.

[27] US Federal Reserve (2026), Financial Accounts of the United States - Z.1, https://www.federalreserve.gov/apps/fof/DisplayTable.aspx?t=f.103.

[4] US Senate, Committee on Banking, Housing, and Urban Affairs (2016), Examining the current trends and changes in the fixed-income markets, (S. Hrg. 114-319), https://www.congress.gov/114/chrg/CHRG-114shrg20948/CHRG-114shrg20948.pdf.

[23] Wu, B. (2020), “Increasing Corporate Bond Liquidity Premium and Post-Crisis Regulations”, SSRN Electronic Journal, https://doi.org/10.2139/ssrn.3613379.

Annex 2.A. Outstanding amounts of corporate debt by country as of end-2025

Copy link to Annex 2.A. Outstanding amounts of corporate debt by country as of end-2025

Annex Figure 2.A.1. Outstanding amounts of corporate debt by country
Copy link to Annex Figure 2.A.1. Outstanding amounts of corporate debt by country

Annex 2.B. Data collection and methodology

Copy link to Annex 2.B. Data collection and methodology

Corporate bond data

Copy link to Corporate bond data

Corporate bond analyses are based on a dataset built by the OECD using deal-level information obtained from LSEG on corporate bond issues that are underwritten by an investment bank. The database provides detailed information for each bond starting in 1980, including e.g. the identity, nationality and sector of the issuer, the type, interest rate structure, maturity date and rating category of the bond, as well as the amount of proceeds obtained from the issue and intended uses thereof.

Convertible bonds, deals that were registered but not consummated, preferred shares, sukuk bonds, bonds with an original maturity less than or equal to one year or with an issue size of less than USD 1 million are excluded from the dataset. Industry classifications are based on The Reference Data Business Classification (TRBC) from LSEG. Annual issuance amounts initially collected in USD were adjusted by 2025 US Consumer Price Index (CPI).

Given that a significant portion of bonds are issued internationally, it is not possible to systematically assign issues to a certain country of issue. For this reason, the country breakdown is carried out based on the issuer’s country of domicile. The advanced/emerging market categories are based on IMF classifications.

The forecast for 2026 issuance used in the infographic is based on the historical average (2000-2025) ratio between gross issuance in year t and the debt coming due in year t (refinancing requirements) as observed at the end of year t‑1. This simple measure performs reasonably well as a predictor of actual issuance, with an average forecast error of 6.6% in the post-2008 period (5.7% excluding 2020, when issuance was significantly above historical trend).

Rating data

Credit rating analyses are based on OECD calculations using data obtained from LSEG. The calculations consider ratings from three leading agencies: S&P, Moody’s and Fitch. For each bond with an available rating in the dataset, the alphanumeric rating is transformed into 21‑point numeric scale with 1 being the lowest rating (C) and 21 the highest (AAA for S&P and Fitch and Aaa for Moody’s). There are 11 non-investment grade categories: five within C (C to CCC+) and six within B (B- to BB+); there are ten investment grade categories: three within B (BBB- to BBB+); and seven within A (A- to AAA).

For bonds with multiple ratings, the average of the available ratings is used. Some bonds do not have rating information available; these are assigned the average rating of all bonds issued by the same company in the same year (t). If the issuer has no rated bonds in year t, year t‑1 and year t‑2 are also considered, respectively. This procedure increases the number of rated bonds in the dataset and hence improves the representativeness of the analyses. When differentiating between investment and non-investment grade bonds, the final rating is rounded to the closest integer and bonds with a rounded rating less than or equal to 11 are classified as non-investment grade.

Early redemption data

Bonds that are no longer outstanding due to having been redeemed before their maturity date are deducted in the annual outstanding corporate bond debt calculations. The early redemption data are obtained from LSEG and cover bonds that have been redeemed early due to being repaid via final default distribution, called, liquidated, put or repurchased. The early redemption data are merged with the primary corporate bond market data via International Securities Identification Numbers (ISINs).

Cost of outstanding debt

The weighted average cost of debt is approximated for fixed-rate debt by calculating the total outstanding amount of debt across 26 coupon buckets between 0 and 1 250 basis points (increasing in increments of 50 basis points, i.e. 0 to 50, > 50 to 100, etc.). The resulting amounts are then multiplied by the midpoint cost of the corresponding bucket (e.g. 25 basis points for the 0 to 50 bucket, 975 basis points for the 950 to 1 000 bucket) and divided by the total outstanding amount of all fixed-rate bond debt. The sum of those products yields an estimated value‑weighted interest rate for all outstanding debt. All debt with coupons above 1 250 basis points are given a midpoint value of 1 275. Where coupon data are not available, the yield to maturity at issuance is used. For bonds that are not issued at par, this may differ from the coupon rate.

Syndicated loan data

Copy link to Syndicated loan data

The syndicated loan analyses are based on OECD calculations using deal-level data from LSEG. This database provides detailed information on each loan, including the borrower’s identity, nationality and sector, as well as the interest rate structure, maturity date and loan amount.

Only loans classified as “syndicated” or “club syndicate” are included in the analysis. Deals with maturities of less than 90 days are excluded. Annual data are based on the closing date, which is when the syndication on all levels/tiers has been signed and completed. Industry-level analyses follow LSEG’s The Reference Data Business Classification (TRBC), while country breakdowns are based on the borrower’s domicile. To account for inflation, issuance amounts originally recorded in USD were adjusted using the 2025 US Consumer Price Index (CPI).

Herfindahl-Hirschman Index of market concentration

Copy link to Herfindahl-Hirschman Index of market concentration

The Herfindahl-Hirschman Index (HHI) is a commonly used measure of market concentration. It is calculated by squaring the market share of each firm active in a given sector and then summing those terms. To analyse concentration in global corporate bond markets, the outstanding amount of debt of each company in the dataset is calculated. That is then divided by the total outstanding amount in the relevant market. The resulting share is squared. This is repeated for each company in the relevant group, after which all the terms are summed up. In general terms:

${H H I}_{s} = \sum_{i = 1}^{N} {(\frac{{O u t s t a n d i n g a m o u n t}_{i, s}}{{T o t . o u t s t a n d i n g}_{s}})}^{2}$

Where Outstanding amount_i,s is the outstanding amount of firm i in sector s, Tot. outstanding_s is the total outstanding amount in that sector and N is the number of firms in the sector.

The HHI can be presented either on a scale from 0 to 1 (if the market shares are given in decimal form) or 0 to 10 000 (if they are given as integers). Commonly used thresholds for what is considered a concentrated market have been established through regulatory guidelines and case law in the field of competition law, where the measure is often used to evaluate the effect that a merger would have on market competition. For example, in its merger guidelines the US Department of Justice and Federal Trade Commission (2024[42]) considers markets with a HHI above 0.1 to be concentrated (“moderately concentrated” if it is between 0.1 and 0.18 and “highly concentrated” if it is above 0.18). The merger guidelines use a 10 000‑point scale which has here been converted to a 0‑1 scale for comparability.

Annex 2.C. Credit spread decomposition

Copy link to Annex 2.C. Credit spread decomposition

This annex provides an explanation of the methodology used to decompose corporate credit spreads. The dataset used is that described in Box 2.1. From this universe three variables area generated: (i) bond-month credit spreads (GZ spreads), (ii) an issuer-level default-risk proxy based on a structural Distance‑to-Default model, and (iii) a bond-level liquidity proxy derived from bid-ask quotes. These are then used to identify the residual, defined as the excess bond premium (EBP). This follows the logic in Gilchrist and Zakrajšek (2012[21]) and Gilchrist et al (2021[22]). After final calculations, each bond’s spread can be divided into a structural part (covering bond characteristics like age, duration, coupons, etc.) and a non-structural component comprising an expected default risk premium, a liquidity premium and the excess bond risk premium. Bond-month portfolio weights are calculated based on the share in the ETF’s portfolio to aggregate bond-level components into a single monthly series. The different steps are covered in turn below.

Bond-month credit spreads: The GZ spread and bond characteristics

Copy link to Bond-month credit spreads: The GZ spread and bond characteristics

A corporate bond spread is the extra yield investors require over a comparable risk-free benchmark. The GZ approach constructs a bond-specific risk-free benchmark by matching the bond’s cashflows to a fitted risk-free zero-coupon curve, and then comparing the bond’s market-implied yield to the yield implied by the risk-free curve for the same cashflows.

Let $y_{b, t}^{c o r p}$ denote the corporate bond yield (defined below) for bond $b$ at month $t$ , computed from its market price. Let $y_{b, t}^{r f}$ denote the cashflow-matched risk-free yield implied by a fitted risk-free curve. Then the GZ spread (in basis points) is:

$s_{b, t} = 10 000 \cdot (y_{b, t}^{c o r p} - y_{b, t}^{r f}) .$

Intuitively:

$y_{b, t}^{c o r p}$ reflects the market price of the corporate bond
$y_{b, t}^{r f}$ reflects what the yield would be if the same cashflows were discounted at risk-free rates
the difference isolates compensation for credit and related premia embedded in the corporate bond price (including related to liquidity, currency, etc.).

Currency consistency

A key requirement for a meaningful spread measure is currency consistency. The corporate yield $y_{b, t}^{c o r p}$ is implied by the bond’s market price and its promised cashflows, and those cashflows are denominated in the bond’s currency. Likewise, the synthetic risk-free yield $y_{b, t}^{r f}$ is constructed by discounting the same cashflows with a risk-free zero-coupon curve, which must therefore be expressed in the same currency.

The risk-free benchmark is built from a US Treasury zero-coupon curve. Therefore, the bond universe used for the GZ spreads (and, by extension, for the EBP estimation) is restricted to bonds denominated in USD. This restriction ensures that:

the yield implied by the dirty price and the yield implied by the synthetic risk-free price are both USD yields, so their difference is economically interpretable; and
the resulting spread does not mechanically absorb currency premia, expected FX depreciation, or cross-currency basis effects that would arise if yields reflected mixed currencies.

Clean prices, dirty prices and accrued interest

Bid and ask quotes as provided by LSEG are treated as “clean” prices. Clean prices exclude accrued interest. However, yields are defined from “dirty” (full) prices.

For each bond $b$ and month-end $t$ a clean mid-price is calculated:

$P_{b, t}^{c l e a n} = \begin{matrix} \frac{B i d_{b, t} + A s k_{b, t}}{2} & if both quotes exist, \\ B i d_{b, t} & if only bid exists, \\ A s k_{b, t} & if only ask exists . \end{matrix}$

Accrued interest is then estimated using the bond’s coupon rate $c_{b}$ (annual, in per cent), coupon frequency $f_{b}$ (payments per year), and a day-count convention. A 30/360 US convention is used to construct a coupon schedule consistent with the bond’s frequency and maturity date. Let $t^{-}$ be the last coupon date before the valuation date $t$ and $t^{+}$ the next coupon date after $t$ . The accrued fraction is then:

$θ_{b, t} = \frac{DC (t^{-}, t)}{DC (t^{-}, t^{+})},$

where $DC (\cdot, \cdot)$ is the 30/360 US day-count. The accrued interest (in price points per 100 par) is:

$A I_{b, t} = \frac{c_{b}}{f_{b}} θ_{b, t}$

Finally, the dirty mid-price is:

$P_{b, t}^{d i r t y} = P_{b, t}^{c l e a n} + A I_{b, t}$

Risk-free benchmark curve and cashflow matching

To construct the cashflow-matched risk-free yield, a fitted government yield curve in the Nelson – Siegel – Svensson (NSS) form (the GSW/Federal Reserve parameterisation) is used. The curve delivers a continuously compounded zero-coupon yield $y_{t}^{0} (τ)$ for maturity $τ$ (in years). Given a sequence of future cashflows ${C F_{b, t, j}}_{j = 1}^{J}$ occurring at times ${τ_{b, t, j}}_{j = 1}^{J}$ , the corresponding risk-free dirty price is computed by exponential discounting:

$P_{b, t}^{r f, d i r t y} = \sum_{j = 1}^{J} C F_{b, t, j} e x p (- y_{t}^{0} (τ_{b, t, j}) \cdot τ_{b, t, j})$

Yield calculation and the “street” convention

Given a dirty price and a fixed coupon $c_{b}$ , the yield $y_{b, t}$ is defined implicitly by a present-value equation. The implementation uses a standard “street” convention with coupon frequency $f_{b}$ that accounts for the fraction $α_{b, t}$ of the coupon period already elapsed at the valuation date. In practice, $α_{b, t}$ is inferred from accrued interest:

$α_{b, t} \approx \frac{A I_{b, t}}{\frac{c_{b}}{f_{b}}}, 0 \leq α_{b, t} \leq 1 f x f x f x f x$

Let $n_{b, t}$ be the number of remaining coupon payments until the bond’s redemption date (maturity or call date; see below). Define the per-period discount factor $1 + \frac{y_{b, t}}{f_{b}}$ . The yield $y_{b, t}^{c o r p}$ solves:

$\begin{matrix} P_{b, t}^{d i r t y} & = \sum_{k = 1}^{n_{b, t} - 1} \frac{\frac{c_{b}}{f_{b}}}{{(1 + \frac{y_{b, t}}{f_{b}})}^{k - α_{b, t}}} + \frac{\frac{c_{b}}{f_{b}} + Par}{{(1 + \frac{y_{b, t}}{f_{b}})}^{n_{b, t} - α_{b, t}}}, \end{matrix}$

where $Par = 100$ for the maturity leg, and $Par$ may differ if the bond is redeemed at a call price.

The same yield-inversion is applied to $P_{b, t}^{r f, d i r t y}$ to obtain $y_{b, t}^{r f}$ that is consistent with the cashflow profile of the bond.

Callable bonds and yield-to-worst (YTW)

Many corporate bonds are callable. Without an option-adjusted spread (OAS) model, a standard practical approach is to work with yield-to-worst (YTW): the yield that is least favourable to the investor among relevant redemption scenarios.

For a callable bond, two yields are computed at each month-end valuation date:

Maturity leg: cashflows until maturity, with principal repayment of 100 at maturity. This yields $y_{b, t}^{c o r p, m a t}$ and $y_{b, t}^{r f, m a t}$ .
Call leg: cashflows until an effective call date with principal repayment at the call price (usually close to 100). This yields $y_{b, t}^{c o r p, c a l l}$ and $y_{b, t}^{r f, c a l l}$ .

In the implementation, the call information (callable flag, call dates, call price) is retrieved from LSEG. Because calls typically occur on coupon dates, the effective call date is defined as the first coupon date that is (i) on or after the first-call date and (ii) strictly after the valuation date. This avoids using call dates that have already passed.

The corporate yield-to-worst is:

$y_{b, t}^{c o r p} = m i n (y_{b, t}^{c o r p, m a t}, y_{b, t}^{c o r p, c a l l}) .$

The matching risk-free yield uses the same leg selection:

$y_{b, t}^{r f} = \begin{matrix} y_{b, t}^{r f, c a l l} & if y_{b, t}^{c o r p, c a l l} < y_{b, t}^{c o r p, m a t}, \\ y_{b, t}^{r f, m a t} & otherwise . \end{matrix}$

Finally, the reported spread for callable bonds is the yield-to-worst spread:

$s_{b, t} = 10 000 \cdot (y_{b, t}^{c o r p} - y_{b, t}^{r f})$

The objective is to obtain a consistent monthly spread measure for a broad universe. Option-adjusted spreads require an explicit interest-rate model and assumptions about volatility and call exercise, which is outside the scope of this implementation. YTW provides a transparent, commonly used approximation when modelling the call option explicitly is not feasible.

Modified duration (maturity and call legs)

Given the characteristics of the bond $b$ , a modified duration measure consistent with the same leg used for YTW can be estimated. The Macaulay duration is

$\begin{matrix} D_{b, t}^{M a c} & = \frac{1}{P_{b, t}^{d i r t y}} (\sum_{k = 1}^{n_{b, t} - 1} \frac{\frac{c_{b}}{f_{b}} ∙ (k - α_{b, t})}{{(1 + \frac{y_{b, t}}{f_{b}})}^{k - α_{b, t}}} + \frac{(\frac{c_{b}}{f_{b}} + Par) ∙ (n_{b, t} - α_{b, t})}{{(1 + \frac{y_{b, t}}{f_{b}})}^{n_{b, t} - α_{b, t}}}), \end{matrix}$

and the modified duration is

$D_{b, t}^{M o d} = \frac{D_{b, t}^{M a c}}{1 + \frac{y_{b, t}}{f_{b}}}$

For callable bonds, $D_{b, t}^{M o d}$ is computed for the maturity and call legs separately and then assigned according to the YTW-selected leg (the same leg selection used for $y_{b, t}^{c o r p}$ ).

From yields to spreads, and to a continuously compounded basis

The raw GZ spread in basis points (bps) is computed as the yield difference between the corporate and risk-free bonds. For the EBP pipeline, spread-type quantities are expressed on a continuous-compounded basis. For an annualised rate $x$ (in bps), the continuously compounded (in bps) equivalent is defined as.

$x_{b p s}^{c c} = 10 000 \cdot l n (1 + \frac{x}{10 000}) .$

This transformation is approximately equal to $x$ for small values but is consistent with exponential discounting and maps simple rates into an additive log-return scale. In the implementation, the spread used in the EBP regression is $s_{b, t}^{c c}$ , as well as the bid-ask liquidity proxy and the coupon-rate control, are expressed on the same continuous-compounded bps scale.

Issuer default-risk proxy: Distance‑to-Default (DD)

Copy link to Issuer default-risk proxy: Distance‑to-Default (DD)

Expected default risk is quantified using an issuer-level, market-based proxy derived from a structural (Merton-style) model. The key idea is intuitive: default is more likely when the market value of the firm (its assets) is close to, or below, an effective debt threshold.

Structural framework and core equations

The Merton model treats equity as a call option on firm assets. Let:

$E_{t}$ : equity market capitalisation (market value of equity) in month $t$
$σ_{E, t}$ : annualised equity volatility
$D_{t}$ : default point (effective debt threshold)
$r_{t}$ : risk-free rate over horizon $T$ (here $T = 1$ year)
$V_{t}$ : (unobserved) market value of firm assets
$σ_{V, t}$ : (unobserved) asset volatility.

Using the Black-Scholes formula, equity is priced as:

$E_{t} = V_{t} Φ (d_{1, t}) - D_{t} e^{- r_{t} T} Φ (d_{2, t}),$

where:

$\begin{matrix} d_{1, t} & = \frac{l n (\frac{V_{t}}{D_{t}}) + (r_{t} + 0.5 σ_{V, t}^{2}) T}{σ_{V, t} \sqrt{T}}, \\ d_{2, t} & = d_{1, t} - σ_{V, t} \sqrt{T} . \end{matrix}$

Given $(E_{t}, σ_{E, t}, D_{t}, r_{t})$ , the implementation solves numerically for $(V_{t}, σ_{V, t})$ by fixed-point iteration using the two equations above, starting from $V_{t} = E_{t} + D_{t}$ and $σ_{V, t} = σ_{E, t} ∙ E_{t} / V_{t}$ , and iterating until convergence. For numerical stability, $σ_{E, t}$ and $σ_{V, t}$ are floored.

Once $(V_{t}, σ_{V, t})$ are obtained, the one‑year Distance‑to-Default is:

$D D_{t} = \frac{l n (\frac{V_{t}}{D_{t}}) + (r_{t} - 0.5 σ_{V, t}^{2}) T}{σ_{V, t} \sqrt{T}}, T = 1 f x f x f x f x$

The implied one‑year expected default frequency (EDF) is:

$E D F_{t} = Φ (- D D_{t})$

Default point $D_{t}$ and cleaning logic

The default point is constructed from balance‑sheet items, recognising that coverage differs across issuers and fields. Three candidate series are used:

$\begin{matrix} D_{t}^{(s t l t)} & = S T_{t} + 0.5 L T_{t}, \\ D_{t}^{(t d)} & = T D_{t}, \\ D_{t}^{(t l)} & = 0.5 T L_{t}, \end{matrix}$

where $S T$ is short-term debt, $L T$ is long-term debt, $T D$ is total debt, and $T L$ is total liabilities.

To avoid switching definitions over time, the script selects one source per issuer based on coverage:

If $D^{(s t l t)}$ has coverage $\geq 0.60$ , use it.
Else if $D^{(t d)}$ has coverage $\geq 0.60$ , use it.
Else use $D^{(t l)}$ if it has any coverage; otherwise pick the source with the highest coverage.

Default-risk measure used in EBP

The EBP pipeline uses a DD-based measure of default risk (no hazard-rate or logit transformations). Specifically, a monotone transformation of DD that increases as default risk rises:

$D e f a u l t P r o x y_{t} \equiv {D D}_{m a x} - {D D}_{t},$

where ${D D}_{m a x} = 20$ in the baseline configuration. This simply shifts and flips the DD scale; it does not change the information content.

Annual fundamentals and monthly carry-forward: Balance‑sheet items used to construct the default point are observed at annual frequency. The monthly panel applies a last-observation-carried-forward rule: each annual observation is assigned to all subsequent month-end dates until the next annual value becomes available. As a result, $D_{t}$ (and hence DD) evolves in a stepwise manner, avoiding the introduction of unobserved within-year dynamics through interpolation. Under this approach, within-year variation in DD is driven primarily by market-based inputs (equity values and equity volatility), while the balance‑sheet component updates only when a new annual report is observed. This treatment is intended for retrospective measurement and descriptive analysis; it should not be interpreted as a real-time estimate of intra-year balance‑sheet evolution.

Safeguards used in the DD pipeline (as implemented)

Structural DD estimates can be sensitive. The implementation therefore applies the below safeguards before constructing the proxy used in the regression.

Safeguard	Setting in code	Rationale
Equity volatility window	12 months (rolling)	Smooths short-lived noise in σE,t.
Volatilities floor	Equity: σE,t≥0.03	Prevents unrealistically low volatility from mechanically inflating DD. Prevents numerical instability in the fixed-point solver.
Volatilities floor	Asset: σV,t≥0.01
Debt outlier filter	Drop if D_t/median outside [1/1000,1000]; or if Dt<107	Mitigates unit errors and extreme miscoding.
Debt staleness filter	Drop if debt age is missing, and keep only if 0 < debt age < DP_STALE_DAYS (to prevent look-ahead); script runs for DP_STALE_DAYS∈{365, 450, 550}	Excludes overly old balance‑sheet values; supports robustness to staleness assumptions (baseline is DPSTALEDAYS=365).
Leverage safeguard (clamp)	If D_t/E_t <0.005, set Dt≔0.005×Et	Avoids spuriously high DD when debt is implausibly small relative to market capitalisation.
DD bounds	DD∈[‑20,20]	Safeguards against bias from tail cases.
Winsorisation of DefaultProxyi,t	1‑99% by month	Limits the influence of extreme tails on downstream regressions.

Liquidity control from bid-ask quotes

Copy link to Liquidity control from bid-ask quotes

Bid-ask spreads capture differences between quoted buying and selling prices and provide a simple proxy for trading frictions. Using the clean bid and ask prices, the liquidity measure (in basis points) is:

$M i d_{b, t}^{c l e a n} = 1 / 2 (A s k_{b, t} + B i d_{b, t}), L i q u i d i t y_{b, t} = 10 000 \cdot \frac{A s k_{b, t} - B i d_{b, t}}{M i d_{b, t}^{c l e a n}}$

The liquidity measure is expressed on the same continuous-compounded bps scale as.

${L i q u i d i t y}_{b, t}^{c c} = 10 000 \cdot l n (1 + \frac{L i q u i d i t y_{b, t}}{10 000})$

computed only when prices are finite, $M i d_{b, t}^{c l e a n} > 0$ , and ${A s k}_{b, t}^{c l e a n} \geq {B i d}_{b, t}^{c l e a n}$ . When liquidity is missing, the liquidity level is set to zero and a dummy variable for missing liquidity is included:

${L i q u i d i t y}_{b, t}^{u s e} = \begin{matrix} {L i q u i d i t y}_{b, t}^{c c} i f o b s e r v e d, \\ 0 o t h e r w i s e, \end{matrix} I_{b, t}^{m i s s} = 1_{{{L i q u i d i t y}_{b, t}^{c c} m i s s i n g}}$

This preserves sample size while allowing missing indicators themselves to carry information.

EBP estimation: Monthly cross-sectional regressions

Copy link to EBP estimation: Monthly cross-sectional regressions

Regression specification and intuition

The dependent variable is the bond-month GZ spread expressed on a continuous-compounded basis point scale, $s_{b, t}^{c c}$ , and the regressors include the issuer default proxy, liquidity, and standard bond characteristics. A pooled OLS regression is estimated over all bond-month observations that pass the data filters:

$\begin{matrix} s_{b, t}^{c c} & = α + γ ∙ D e f a u l t P r o x y_{b, t} + λ ∙ L i q u i d i t y_{b, t}^{u s e} + δ ∙ I_{b, t}^{m i s s} \\ + β_{1} ∙ l n (D_{b, t}^{M o d}) + β_{2} ∙ l n (T T M_{b, t}) + β_{3} ∙ l n (A g e_{b, t}) + β_{4} ∙ {C o u p o n}_{b, t}^{c c} + u_{b, t}, \end{matrix}$

where, for the bond $b$ , $T T M_{b, t}$ and $A g e_{b, t}$ are the time‑to-maturity and time‑since‑issuance (in years), respectively, ${C o u p o n}_{b, t}^{c c}$ is the coupon rate expressed in continuously compounded bps, and the coefficients are estimated by ordinary least squares.

DefaultProxy: isolates the component of spreads related to expected default risk, measured continuously and updated monthly.
Liquidity: accounts for trading frictions that can widen quoted spreads even when default risk is unchanged.
Bond characteristics (duration, time to maturity, age, coupon): control for predictable cross-sectional variation in spreads linked to maturity structure and interest-rate sensitivity.

After controlling for these observable components, the residual $u_{b, t}$ captures the remaining part of the spread, interpreted as an “excess” premium (risk compensation and other unobserved frictions not captured by the controls), i.e. the EBP.

Definition of bond-level EBP and monthly aggregation

The bond-level EBP is the regression residual:

${\hat{E B P}}_{b, t} \equiv {\hat{u}}_{b, t} = s_{b, t}^{c c} - {\hat{s}}_{b, t}^{c c} .$

The monthly EBP series is the value‑weighted average of residuals:

$E B P_{t} = \sum_{b \in B_{t}} w_{b, t} ∙ {\hat{E B P}}_{b, t}$

The corresponding value‑weighted spread and fitted components are:

$S p r e a d_{t} = \sum_{b \in B_{t}} w_{b, t} ∙ s_{b, t}^{c c}, F i t t e d_{t} = \sum_{b \in B_{t}} w_{b, t} ∙ {\hat{s}}_{b, t}^{c c},$

so that (up to rounding and missing-value handling) $S p r e a d_{t} \approx F i t t e d_{t} + E B P_{t}$ .

Weights used for aggregation

Weights $w_{b, t}$ are built from the representative weight in the ETF for each bond and normalised to sum to one within each month in case of missing data. This ensures the reported EBP reflects the composition of the underlying ETF-based bond universe.

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment

Governance

Health

Industry, business and entrepreneurship

Regional, rural and urban development

Science, technology and innovation

Society

Taxation

Trade

Energy

Nuclear energy

Transport

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Countries

Countries A - C

Countries D - I

Countries J - M

Countries N - R

Countries S - T

Countries U - Z

Regional and global engagement

Publications

Publications

Featured publications

Data

Data

Featured data

News & events

News & events

Featured events

About OECD

About

Engage with us

Work with us

Featured topics

Agriculture and fisheries

Climate change

Development

Digital

Economy

Education and skills

Employment

Environment

Finance and investment