Empiricalanalysis

In the rest of this section we present, based on AABM, some results from trying to test the above predictions.

2.2.1 Data and measurement

Annual growth is computed as the log difference of per capita income obtained from the Penn World Tables (PWT) mark 6.1.

Financial development is measured by the ratio of private credit, that is the value of loans by financial intermediaries to the private sector, over GDP. Data for 71 countries on 5-year inter­val averages between 1960 and 1995 (1960-4, 1965-9, etc.) was first compiled by Levine et al. (2000); an annual dataset was more recently prepared and made available by Levine on his webpage. Private credit is the preferred measure of financial development by Levine et al. because it excludes credit granted to the public sec­tor and funds coming from central or development banks. AABM also conduct sensitivity analysis with two alternative measures of credit constraints: liquid liabilities and bank assets. The first is defined as currency plus demand and interest-bearing liabilities of banks and nonbank financial intermediaries divided by GDP; the second gives the value of all loans by banks but not other financial intermediaries.

The next step in presenting the evidence is to look at the response of growth to specific shocks. AABM first consider terms of trade shocks, available as 5-year averages between 1960 and 1985 from the Barro-Lee (1996) dataset. Changes in the terms of trade reflect export-weighted changes in export prices net of import-weighted changes in import prices, quoted in the same currency. Arguably, exchange rate fluctuations may be endogenous to the growth pro­cess and therefore regressions of growth on terms of trade shocks may be subject to reverse causality and produce biased coeffi­cient estimates.

For the analysis on the transmission channel of credit con­straints, one also needs data on long-term versus short-run investments. AABM consider R&D as a share of total investment. Unfortunately, data availability limits the sample to 14 OECD countries between 1973 and 1997 for which the OECD reports spending on R&D in the ANBERD database. Data on investment as a share of GDP is easily obtainable from the PWT.

In the growth regressions AABM follow Ramey and Ramey and Levine et al. in controlling for population growth, initial secondary school enrollment, and a set of four policy variables (the share of government in GDP, inflation, the black market exchange rate premium, and openness to trade). We use demographics data from the PWT and the policy conditioning set in Levine et al.

2.2.2 AsummaryoftheAABMresults

While Ramey and Ramey study the response of long-term growth to volatility and Levine et al. focus on the direct effects of credit constraints on growth, our model predicts that volatility is more harmful to long-run growth in financially underdeveloped countries. AABM therefore estimate the basic equation:

where yi is the initial income in country i, gi denotes the aver­age rate of productivity growth in country i over the whole period 1960-95, voli is the measure of aggregate volatility, priv_i- is the aver­age measure of financial development over the period 1960-95, Xi is a vector of country-specific controls, and ε_t is the noise term. We are mostly interested in the interaction term α4 ∙ vol_i- * priv_i-, and our first prediction is that α4 should be positive and significant.

Table 2.1 presents the results reported in AABM. They find a strong direct negative effect of volatility on long-term growth of -0.41 and a significant positive coefficient on the interaction term of 0.011 (Column (1)). In this sample private credit varies from 4% to 141%, with a mean of 38% and a standard deviation of 29%. A one standard deviation increase in the level of financial develop­ment would therefore reduce the impact of a 1% rise in volatility from a 0.41% fall in the growth rate to -0.41 + 0.011 * 29 = -0.09%. This effect is significant in economic terms not only because of the large variation in private credit in the cross-section, but also because of its substantial fluctuations in the time series: For example, in the United States private credit almost tripled between 1960 and 1995, steadily rising from 50% to 140%. For many coun­tries the level of private credit moved up and down significantly during the same period. As Column (2) shows, the AABM result is robust to the inclusion of demographic and human capital con­trols, as well as measures of property rights protection and the policy variables from Levine et al.

For sufficiently high levels of private credit (which we observe for many OECD countries) these results predict that the overall contribution of volatility to economic growth becomes positive. Moreover, for intermediate levels of private credit the gross con­tribution may be close to zero. Regressing long-run growth on

Table 2.1 Growth, volatility, and credit constraints: basic specification

No investment

With investment

Whole sample OECD countries Whole sample OECD countries

Note: Dependent variable is average growth over the 1960-95 period.

volatility alone without accounting for the direct and interacted effects of financial development, could thus produce an insigni­ficant coefficient. This may explain why Ramey and Ramey find a strong negative effect of volatility on growth in the full cross­section but a nonsignificant one in the OECD sample. In Columns (3) and (4) AABM estimate the above equation for the OECD coun­tries only, and find coefficients similar to the ones we find for the entire sample.

Finally, Columns (5) and (6) show that the growth impact of both volatility itself and its interaction with private credit are little affected by the inclusion of investment as a control. Risk arguably affects savings rates and investment, and investments fuel growth. However, controlling for the ratio of investment to GDP reduces the coefficient on volatility by only 20%, suggesting that 80% of the total effect of volatility on growth is via a channel other than the rate of investment.

These results have been shown to be robust to alternative measures of financial development, namely liquid liabilities and bank assets.²

Next, to address the potential endogeneity problem raised by the above measure of volatility, AABM analyze the sensitivity of growth to exogenous shocks, exploring both the cross-section and time-series variation in the panel. They consider average data over 5-year period intervals in a cross-section of over 70 countries to estimate the following specification:

where git and privy are the annual growth and private credit aver­ages for country i in the 5-year subperiod t, and yu is the initial per capita income at the beginning of the subperiod.

Table 2.2 The response of growth to terms of trade and commodity price shocks: 5-year averages

Note: Dependent variable is average growth over 5-year intervals in the 1960-85 period.

Source: AABM (2004), table 4.

Table 2.3 The response of investment to commodity price shocks: annual panel data, fixed effects

Controls

Note: Dependent variable is investment as a share of GDP or R&D as a share of investment. Annual 1973-97 data, except where lost due to lags. Panel fixed effects estimation. Shockt, shockt-1, shockt-2 refer to the contemporaneous, 1-year and 2-year lagged commodity price shock, as defined in the text. Lagged (t-10,t-6) or (t-5,t-1) average used for private credit, as indicated in the column heading. All regressions include a constant term and a linear trend. T-statistics in parenthesis. ***,**,* significant at the 1%, 5%, and 10% respectively.

Source: AABM (2004), table 7.

also be positive. We are particularly interested by the interaction term α4 ∙ shock_i'_t * priv_rt and we predict that α4 should be negative, which says that a higher level of financial development will imply a lower sensitivity of growth to shocks. In the estimation we allow for country-specific fixed effects.

Table 2.2 summarizes the results from AABM. In particular, it shows evidence of a significantly positive effect of shocks on medium-run growth. We also observe a strong negative coeffi­cient on the interaction term, although it is only significant when we consider changes in the terms of trade. This result is shown to be robust to alternative measures of financial development and to using a one-period lagged value of private credit. Because of the substantial time-series variation in private credit it is not sur­prising that using its initial 1960 value produces an insignificant interaction term.

The final link that remains to be established is the interaction of financial development with the cyclicality of long-term investment and total investment with respect to exogenous shocks. As a proxy for long-term investment, AABM use data on R&D expenditures. Using annual data on 14 OECD countries between 1973 and 1997, AABM perform two first-stage regressions, respectively for R&D as a share of total investment (R&Dn/In) and investment as a frac­tion of GDP (²ö/GDPn) for country i in year t, as a function of current and lagged shocks and current and lagged shocks (one and two period lags) interacted with private credit in the coun­try during subperiod t. As before, they include a linear trend and allow for country fixed effects.

Table 2.3 displays the results from these two regressions. Column (5) uses the moving average of private credit over the immediately preceding 5 years. In line with the model, we see that positive shocks stimulate R&D for a given level of total invest­ment, but the ratio of R&D over total investment tends to be countercyclical in the absence of credit constraints. In particular, all coefficients on lagged shocks are positive, while the interac­tion terms of these shocks with private credit have a negative and significant coefficient. Two-year lagged shocks are the only sig­nificant ones, suggesting that the reallocation of investment itself takes place with a lag. To gauge the importance of credit con­straints, note that R&D is procyclical for low levels of financial development but a value for private credit of 0.74/1.25 = 60% is enough to make long-term investment countercyclical with respect to twice-lagged shocks. In fact, we observe such high levels of loan availability for many countries in our sample of 70, with the num­ber tripling to 36 between 1974 and 1999. These results are shown to be robust to using the 5-year lagged 5-year average of private credit (Columns (6)-(8)).

In contrast to the above findings, the share of total investment in GDP does not become more procyclical as credit constraints tighten (Table 2.3, Columns (1)-(4)). If anything, financial develop­ment may magnify the procyclicality of the investment over GDP ratio, if we rely on the only significant (and positive) interaction coefficients in Columns (2) and (4). Overall, these results support our theory that credit availability redirects resources toward long­term projects (such as R&D) during a downturn, translating into improved growth a year or two later.

2.3