The drivers and impacts of subsidies to steel firms

Annex E. Explaining subsidies: estimations results

We used both Pseudo-Poisson Maximum Likelihood (PPML) estimations and linear regressions to assess the impact of firms’ characteristics on their provision of grants and BMB. PPML have been successfully used in the context of explaining subsidies in to (Branstetter, Li and Ren, 2023[18]), and allow the explained variable to take 0 values, thus keeping observations of firms which received no grants nor BMB on a given year in the sample used for estimation.

All explanatory variables except ratios (indebtedness, EBITDA, capacity utilisation, etc.) and dummy variables are in log-form. Furthermore, all explanatory variables are lagged by one year.

Results are reported in the following tables. “+” indicates a statistically significant positive coefficient whereas “-“ indicates a statistically significant negative coefficient. The symbol . indicates no statistically significant impact. When variables of no statistical significance were found, they were then withdrawn one by one starting with the one of highest p-value, and the model was re-assessed, to ensure statistically significant variables remained so, until only statistically significant variables were left. Precise coefficient can be provided on demand.

As can be seen in the tables below, impact of statistically significant variables in PPLM estimations are often corroborated by the results of multivariate linear regression (LR). When the variable of business cycle (ln_GDP) was not significant, year effects were introduced, often yielding similar results hence not changing any of the conclusions.

Table 3. Correlations to grants, grants intensities, BMB, and BMB intensities

Pseudo-Poisson Maximum Likelihood (PPML) estimations and linear regressions results

grants	PPML, All sample	Linear, All sample	PPML, OECD	Linear, OECD	PPML, partner	PPML, partner	Linear, Partner economies
asset	+	+	-	.	+	+	+
debt/asset	.	.	-	-	+	.	.
revenue	-	-	+	+	(vif too high)		-
% owned by government	+	+	.	.		+	+
EBITDA	+	+	.	.	+	+	+
GDP_steel	+	.	+	+	.	.	.
gross profit margin	.	.	.	.	.	.	.
cap_util	.	.	.	.	.	.	.
grants/asset	PPML, All sample	Linear, All sample	PPML, OECD	Linear, OECD	PPML, partner economies		Linear, partner economies
debt/asset	+	.	-	. / -	+		.
soe_perc	+	+	-	. /	.		.
EBITDA	.	+	.	. /	.		+
GDP_steel	+	.	+	+ /+	+		.
gross_profit_margin	.	.	.	. /	.		.
cap_util	+	.	.	. /	+		.
BMB	PPML, All sample	Linear, All sample	PPML, OECD	Linear, OECD	PPML, partner economies		Linear, partner economies
asset	+	+	.	.	+		+
debt/asset	+	+	.	.	+		+
revenue	.	.	.	.	+
soe_cat	+	+	+	+	+		+
EBITDA	.	+	.	.	-		+
GDP_steel	.	.	.	.	.		.
gross_profit_margin	.	.	.	.	.		.
cap_util	.	.	.	.	+		.
BMB/asset	PPML, All sample	Linear, All sample	PPML, OECD	Linear, OECD	PPML, partner economies		Linear, partner economies
debt/asset	+	+	+	+	+		+
soe_perc	+	+	+	+	+		+
EBITDA	-	.	.	.	-		.
GDP_steel	+	.	.	.	.		.
gross_profit_margin	.	.	.	.	.		.
cap_util	+	.	-	-	+		+

Note: all explanatory variables in lagged log form, except for dummy variables and ratios; SOE_per is not significant in linear regressions if you had country dummy, for which only LUX and CHN are significant. For PPLM it is still significant after adding country dummy yet the coefficient becomes decreasing.