Further Statistical Debate On ‘Too Much Finance’
Peterson Institute for International Economics
October 23, 2015
This paper evaluates recent findings by researchers at the Organization for Economic Cooperation and Development (OECD) on “too much finance.” It first critiques the OECD findings, which seem to imply that the optimal amount of finance is zero, given the linear specification of the main tests. It then finds that the negative impact of additional finance on growth is reversed when the appropriate (purchasing-power-parity) per capita income is applied and country fixed effects are removed. Separate tests for countries with intermediated finance below and above 60 percent of GDP show a significant positive effect of finance on growth in the lower group but an insignificant effect in the higher group. An appendix replies to critics of my earlier study (Cline 2015b) in which I argued that an estimated negative quadratic effect of finance on growth was likely to be a spurious correlation reflecting convergence-based lower growth at higher per capita incomes. It notes that the critics’ own logarithmic tests, yielding a positive marginal impact of finance on growth even at high levels, achieve comparable explanation to their quadratic form yielding a negative marginal impact. It finds that adding dummy variables for below and above intermediate financial depth to the logarithmic form does not support the inverse U influence found in the quadratic form.
Further Statistical Debate On ‘Too Much Finance’ – Introduction
In a recent PIIE Policy Brief, Too Much Finance or Statistical Illusion? (Cline 2015b), I question recent statistical findings that financial depth has reached an excessive level in advanced economies and has contributed to a slowdown in growth as a consequence. Some studies have found that whereas a rising ratio of private credit to GDP initially spurs growth, once this ratio exceeds about 100 percent additional finance begins to reduce the growth rate. Th e key innovation in these studies is to add a quadratic term to the finance variable, and the key fi nding is that the quadratic term is negative so that the growth contribution of finance rises but then falls again after a peak (Arcand, Berkes, and Panizza 2012; Cecchetti and Kharroubi 2012; Sahay et al. 2015).1 My critique of these studies is that there is an inherent mathematical bias toward finding a negative quadratic term on finance if there is a positive linear term and at the same time there is a positive trend relationship between financial depth and per capita income. I demonstrate this bias by running statistical tests that also find a significant negative quadratic term if growth is “explained” by doctors per capita, or R&D researchers per capita, or the number of telephones per capita. We should have no more confidence in the finding that too much finance spoils growth than in the finding that too many doctors, R&D researchers, or telephones spoil growth. Arcand, Berkes, and Panizza (2015b) have issued a comment on my analysis. Appendix A sets forth my reply to their comment.
Now a new study from another international financial organization has arrived at statistical findings with even more astonishing implications: For OECD countries, additional financial depth uniformly reduces growth.2 In the main results of two researchers at the OECD (Cournède and Denk 2015), cross-country growth equations show a strictly negative coefficient on a linear variable for financial depth (ratio of private credit to GDP) without any quadratic term. If these results were taken literally, there would be a radical policy implication: Growth would be maximized by completely eliminating credit finance. The optimal amount of credit would be zero. Th e authors do conduct supplementary tests that suggest the influence of finance on growth is positive at initially low levels of finance, and this enables them to state in their abstract: “…finance has been a key ingredient of long-term economic growth in OECD and G20 countries over the past half-century….” But they then assert that “at current levels of household and business credit further expansion slows rather than boosts growth” (p. 3). However, they seek to support this conclusion on the basis of the strictly linear negative coefficient that is estimated for the full sample, including low-financial-depth observations, and that test inescapably implies that the optimal level of finance is zero. Th e authors cannot reject the fully linear results when it comes to the implication of optimal zero finance but at the same time use them as the basis for asserting that at “current levels” the impact of additional finance on growth is strictly negative. The authors’ (proper) insistence on a positive growth influence of finance over an initial range seriously undermines the usefulness of their main estimates.
More fundamentally, even if attention is restricted to a range of private credit above say 60 percent of GDP, there is a major problem regarding causality. Higher per capita income is likely to drive relatively more demand for credit as, in effect, a luxury good. If so, when combined with the long-recognized “convergence” pattern of lower growth at higher per capita incomes, the effect will be that higher credit is observed to accompany lower growth but without causality. Reduction of credit would thus not boost growth because high credit is not causing low growth; instead, the maturing of the economy is slowing growth.
It turns out, moreover, that the study’s main statistical finding does not hold up to certain key changes in specification. Th is working paper uses the same dataset as Cournede and Denk (2015), kindly provided by the authors, to examine this question.3 Th e central findings here are that the results of that study are unreliable because first, the tests exclude the most important variable, real per capita income at purchasing power parity (ppp) comparable across countries; second, the tests apply country fixed effects and thereby throw out important information on cross-country variation; and third, incorporation of shift and slope dummy variables for lower financial depth removes the significance of the negative influence of higher financial levels on growth while tending to confirm the expected positive influence at low levels.
The Cournede-Denk Results
Cournede and Denk use data for 33 OECD countries for the period 1961–2011 to estimate cross-country growth regressions incorporating a linear term on finance. I will focus on their results for “intermediated credit,” credit to the private sector from either banks or other financial institutions.4 Table 1 reports their coefficient estimates for three specifications of regressions for per capita GDP growth. In the simplest, only the credit variable is included. It has a significant negative sign.5 In the second, once again annual data are applied, but other variables are added: the investment rate, average years of schooling, population growth rate, and a dummy variable for banking crisis. Again the credit variable has a significant negative coefficient. In the third variant, data are grouped into five-year averages. Once again the credit variable has a significantly negative impact.
See full PDF below.