Econintersect: Two IMF economists have posted a working paper which attempts to define relationships between the presence of state owned banks, the occurrence of higher public debt and lax fiscal discipline. The authors, Jesus Gonzalez-Garcia and Francesco Grigoli, are responsible for the content which “do not necessarily represent those of the IMF or IMF policy“. In the abstract (following the Read more >> break), it is said that the authors have found state-owned banks associated with high public debt, larger fiscal deficits and crowding out of public sector credit, among several other presumed negative characteristics.
China’s ICBC state-owned bank, the world’s largest bank.
Here is the abstract (with emphasis added by Econintersect):
State-owned banks may help to soften the financing constraints of public sector entities and consequently become a factor that hampers fiscal discipline. Using a panel dataset, we find that a larger presence of state-owned banks in the banking system is associated with more credit to the public sector, larger fiscal deficits, higher public debt ratios, and the crowding out of credit to the private sector. These results suggest that the lending practices of state-owned banks should be carefully assessed in any strategy to pursue fiscal discipline.
The authors start (right after the Introduction) with a fundamental statement about state-owned banks, including a definition (emphasis added by Econintersect) of the operational mechanism of commercial banking:
There are four groups of state-owned financial institutions: retail commercial banks, development banks, quasi-narrow banks, and development agencies.
Retail commercial banks generally have social or development objectives but carry out thesame type of operations as private commercial banks. In particular, they collect deposits and use them to give credit to firms and individuals, hence acting as first-tier banks on both the asset and liability sides.
Read the entire paper below. Econintersect commentary follows the Scribd panel.
Econintersect commentary
There are a number of questions that we feel the authors need to address:
- The authors have an improper definition of the structure of commercial banks operations. They are presenting a “loanable funds” definition of bank operations which is not widely accepted today. How central is this definition to their model?
- There is appears to be no consideration given in the modeling to the differences between states that use a currency not issued by themselves (or have a fixed peg to another currency) and those who issue their own sovereign currency. Examples of the former are all countries in the Eurozone (EU-17) and Argentina for several years in the 1990s. Examples of the latter are the 10 EU-27 countries not in the EU-17, Brazil, India and Russia. Are there any differences in data patterns when such distinctions are introduced?
- The authors’ data comes from “…the 96 countries where the government owns some commercial banks, its ownership of the banking system is sizable.” (Authors’ Fig. 1) The two largest banks in the world are in China (ICBC and China Construction Bank), the sixth largest bank is Agricultural Bank of China and the ninth largest is (Bank of China). All are state owned banks. Yet China does not appear on the list of countries with government ownership of the banking system. They have omitted the world’s second largest economy in which 75% of the banking is state-owned.
- The authors claim that
“Figure 2 … shows a positive and significant relationship between government’s participation in the banking system and credit to the public sector as share of total assets.”
Red annotations added by Econintersect.
Questions that arise from inspection of Fig. 2:
- There is no correlation coefficient given for the regression line. In Econintersect experience we would expect the vallue of R-squared to be very small, possibly close to or less than 0.20 which we generally consider to be ‘negligible correlation’.
- The three data points within the red box appear to be outliers. Have they been tested to determine that they actually are within the same distribution as the rest of the data points?
- What does the regression line look like (and how does r-squared change) if the three outlier data points are excluded?
- The two highest points on the y-axis (zero state ownership of banks) that appear to be slightly higher in y-value than does the highest percentage of state bank ownership. If there is a significant relationship between x and y in this plot discussion of the particular circumstances of those three (and other nearby) data points might prove instructive. (Of course the suspected very low correlation coefficient does call into question whether there is a significant relationship.)
- If the regression line drawn is accepted as correct, then, on average, moving from 0% to 75% state ownership of banks produces, on average, an increase of public sector share of total assets from about 9% to approximately 28%. This is lacking context without some understanding of how total credit varied across the bank ownership spectrum (normalized against some reference such as GDP). Did total credit as a percentage of GDP increase going from left (0% state-owned banking) to right? Did it decline or remain the same? The interpretation of the change from 9% to 28% will depend on those answers.
- The authors state:
“Figure 3 shows no discernible relationship between government’s ownership of the banking system and the overall balance. These banking system and budget deficit nexus, however, should be investigated in a multivariate regression framework to control for factors that may influence these relationships.”
Red lines added by Econintersect.
As for Fig. 2 the absence of a correlation coefficient for a scatter graph showing wide variations does not lend confidence to the usefulness of the regression fit.
There is a pronounced narrowing of the vertical scatter of data point (marked by the two red lines). While in might be tempting to ascribe a narrower degree of government debt variability with increasing state ownership of banks, a statistician should be consulted before any confidence could be given to that observation. The authors did not discuss this observation as far as we have determined so far. It was unrelated to the hypotheses they chose to test.
The authors took this trendless data and subjected it to another covariant model analysis to come up with a conclusion which is not evident in the raw data scatter graph:
“…the estimates suggest that each additional percentage point in the share of banking system’s assets owned by the governmentis associated with an increase in public debt of 0.2 – 0.3 percent of GDP.”
- The authors state that:
“The possibility of crowding out of credit to the private sector can be inferred from the resultsin Section III.A.”…”This negative relationship between government ownership of banks and credit to the private sector is clearly depicted in Figure 4 and is statistically significant.”
Here as with Fig. 2 there is no correlation coefficient given for the regression line. The significant scatter suggests a low R-squared value again. If one starts with a poor or insignificant correlation, how can one have confidence in further modeling? If there is a basis for this the authors should provide more information.
If the line as drawn is accepted, then going from 0% state ownership to 75% will, on average, lower the private sector share of total assets from about 85% to approximately 70%. Just as we discussed above for Fig. 2 this is a meaningless observation unless the context of how total assets are changed (on average) in going from 0% to the 75% extreme for state ownership
- The authors state:
Other significant regressors are GDP growth, trade openness, and oil exports. The negative relationship with GDP growth suggests that more dynamic economic activity is associated with a higher fiscal deficit. While this seems counter-intuitive,…”
Econintersect wonders why it is counter-intuitive that expanding credit can correlate with positive GDP growth?
While the authors of the current paper have a lot of work to do to present a supportable analysis, it would appear they are addressing a potential audience ready to accept their propositions with whatever analysis they present. Many academics have held the view for some time that state-owned banks are inefficient pariahs. A particularly good essay on this point is by Charles Calomiris, Henry Kaufman Professor of Financial Institutions at Columbia University. (Note: There is another essay at the same website as Calomiris posted which is supportive of the value state-owned banks.)
The authors of the current paper provide an extensive bibliography of supportive research.
It appears that many in the world think that they are correct. But that cannot be an excuse to present less than thorough research. And Econintersect remains suspicious of any covariant model analysis that succeeds in producing trends that are not evident in the raw data.
Important note: the authors repeatedly use the words “associated with“. We will not be surprised to find the work cited as showing one thing causes another, which will be unfortunate. Hopefully these researchers will do what some others have not done previously and that is to keep going on record that they have shown “associations” and not “causations“. From the Econintersect point of view, it is also fortunate that they chose not to use the term “correlation” for any of their observations. As we have discussed we do not think they have shown any correlations.
Of course, this is a working paper in quest of comments – and we have been happy to comply.
Sources:
- State-Owned Banks and Fiscal Discipline (Jesus Gonzalez-Garcia and Francesco Grigoli, IMF Working Paper WP/13/206, October 2013)
- World’s Largest Banks (Banks Around the World)
- The AAF Virtual Debates: Charles Calomiris on State-Owned Banks (Charles Calomiris, All About Finance, blogs.worldbank.org, 02 February 2011)
- The AAF Virtual Debates: Franklin Allen on State-Owned Banks (Frankiln Allen, All About Finance, blogs.worldbank.org, 02 February 2011)