Analysis Of The Relationship Between Debt-Gdp Ratio And The Growth Rate

Intro:

This paper provides a summary of the important literature surrounding the relationship between debt-GDP ratio and the growth rate. The authors find holes in each method, but come to the conclusion that a potential relationship between debt and GDP should still be pursued further. However, the 90% threshold shouldn’t be considered the be all and end all for governments; each country has different specifics that alter how debt and GDP reacts.

Ways to Measure Data:

A key point that must be defined early in any of the studies is what definition of debt they are using. “Gross government debt measures the stock of outstanding government debt and net government debt is the difference between gross debt and the financial assets held by the government” (Panniza and Presbitero, 2013). The difference between the two can be very large. The authors state that net debt is best to use, although it’s very difficult to calculate accurately; each country will have a different definition of a ‘financial asset’, making cross country analysis difficult (Panniza and Presbitero, 2013).

Theory:

The authors “assume that government expenditure in goods and services is fixed and we examine what happens if the government decides to temporarily reduce taxes and finance its expenditures by issuing debt. ” (Panniza and Presbitero, 2013). In the short-run, an increase in fiscal spending (funded by government debt) will lead to an increase in AD, leading to an increase in real GDP. However, in the long-run, private savings doesn’t match the government spending, and so there’s less funds for borrowing. This leads to a decrease in investment, and a fall in the GDP growth rate. Basic calculations show that a 100% increase in the debt levels according to GDP would lead to a 20% decrease in the GDP growth rate (Panniza and Presbitero, 2013).

Many studies have determined that an inverse-U relationship between debt and GDP growth is present. However, there’s no theoretical model that would explain this. Therefore, the entire relationship between the two could be a cause of correlation, not causation. To summarize, Panniza and Presbitero, (2013) believe that “debt may have a negative effect on growth, but the effect is likely to be small”. The effect is also very likely to depend on many factors: the overall state of the economy, the amount of debt that was issued, and the composition of the debt, amongst many things.

Empirics:

Minea and Parent (2012) use a panel smooth threshold regression model to model the relationship. They find that a negative effect is seen when the debt to GDP ratio is between 90 and 115% of GDP. The presence of an upper-bound is interesting and hasn’t been replicated in many studies. (Minea and Parent, 2012)

Many of the studies also use a 5 year panel to try and counteract for covariates between debt and growth. However, this limits the amount of data points that are available for analysis. Therefore, the studies use a 5 year overlapping period, but this causes autocorrelation (Panniza and Presbitero, 2013).

Almost all the studies struggling with endogeneity. To control for endogeneity, you can use a model where debt is a function of growth and growth is also a function of debt (a bi-variate model), and then calculate OLS (Panniza and Presbitero, 2013). Using internal variables (such as lagged values of debt and GDP growth rates) can be used along with GMM estimation. However, “GMM estimators were developed for micro data and are poorly suited for macroeconomic data sets with a relatively low number of cross-sectional units” (Panniza and Presbitero, 2013). External instruments can be used but it is difficult to establish exogeneity with them. However, often, the 2sls estimates and the OLS estimates are very similar, meaning that either debt isn’t endogenous (highly unlikely), or that instrument used is poor.

Most of the papers establish that there’s evidence of a non-linear relationship between debt and GDP growth. The easiest way is to include a quadratic term, like Checherita and Rother (2012) have done, but this method is heavily affected by outliers (Panniza and Presbitero, 2013). Alternatively, spline regression can be used (dived the data into portions and try to fit a line of best fit through each region).

Panel threshold regression can be used to estimate the thresholds in each model. This allows a more precise measure, rather than segment (i. e 60-90% of GDP). This method was established by Hansen(1996). However, when past papers have used this method, key assumptions of the method have been violated, and not accounted for (e. g Hansen assumed the errors are i. i. d. , which isn’t true, and it hasn’t been established whether the method works with heteroskedastic errors) (Panniza and Presbitero, 2013). Also, the non-linearities present in the data are probably a lot more complicated that what is accounted for in previous studies (Panniza and Presbitero, 2013).