Empirical Analysis of Policy Responses and Recovery

Most of this section's analysis is borrowed from Hong, Lee, and Tang (2010b).

3.1 Data

In my analysis, I investigate 21 developing Asian economies and 21 industrialized economies. While most previous studies on recessions and recoveries consider only developed economies (Claessens et al. 2008; IMF 2009a), this paper examines Asian developing economies as well to provide a more useful comparison with the Korean economy. The developing Asian economies in my sample include: the 10 Association of South East Asian Nations (ASEAN);⁴ Bangladesh; People's Republic of China (PRC); Hong Kong, China; India; Kazakhstan; Korea; Kyrgyz Republic; Pakistan; Papua New Guinea; Sri Lanka; and Uzbekistan. The industrialized economies refer to 21 Organization for Economic Co-operation and Development (OECD) countries, namely Australia; Austria; Belgium; Canada; Denmark; Finland; France; Germany; Greece; Ireland; Italy; Japan; Netherlands; New Zealand; Norway; Portugal; Spain; Sweden; Switzerland; United Kingdom; and US. The sample period is from 1961 to 2008. The data set is unbalanced due to missing observations.

In dating recession periods, I use annual, not quarterly, real GDP figures from World Development Indicators and OECD. Quarterly GDP series for developing economies are very limited and not seasonally adjusted in most cases. For analyzing policy responses, we use changes in government consumption expenditure as a measure fiscal policy and changes in the call interest rate as a measure of monetary policy. Fiscal policy response is my main interest and monetary policy is considered mainly for controlling purposes. I use government consumption expenditure instead of a more standard measure such as primary balance because the latter has very low data availability, especially among developing Asian economies.⁵ Both the call rate and government consumption are obtained from the International Monetary Fund's International Financial Statistics and OECD. The call rate is adjusted for inflation, where inflation is defined as annual growth in consumer price inflation.

3.2 Definitions of Recession and Recovery

I first date peaks and troughs in GDP series based on the concept of "classical" business cycles which was formalized by Burns and Mitchell (1946) and later implemented in algorithm by Bry and Boschan (1971) and Harding and Pagan (2002). While relatively simple, this dating practice is known to closely match the business-cycle dates provided by the National Bureau of Economic Research (in the case of the United States) or other more complicated approaches. When applied to an annual series, Harding and Pagan's rule implies that period t should be defined as a trough if xt < x_t+k, for k = -1 and 1, and a peak if xt > x_t+k, for k = -1 and 1. In other words, any year of negative growth can be regarded as a recession or a downturn. Watson (1994) uses the same reasoning in his analysis of annual data. I apply this rule to my annual real GDP series.⁶ Once peaks and troughs are dated, a recession is naturally defined as the period that lies between a peak and the following trough. The duration is simply given by the length (number of years) of a recession, and the amplitude by the peak-to-trough fall in the logarithmic value of the series. According to these definitions, the sample probability of recession is 8.4% for OECD and 9.9% for developing Asia. Also, the average duration of a recession is 1.3 years for OECD and 1.5 years for Asia, while the average amplitude is -1.7% for OECD and -6.3% for Asia. For the Korean economy, the recession probability is only 4%: Korea has experienced only two recession episodes during the whole sample period and each one lasted for one year. The amplitude of Korea's recession was -4% on average.

It is less straightforward to define a recovery, because people seem to use the word “recovery” to imply many different things. In this paper, I use the following two definitions: (1) the time length until recovery to previous peak, and (2) the growth rate during the first year of recovery. The first definition measures how long GDP, after hitting a trough, remains below the previous peak. The second definition measures the GDP growth rate during the one year period that immediately follows a trough. These are the same definitions as those used in World Economic Outlook by IMF (2009a). Figure 4 [ PDF 22.5KB | 1 page ] illustrates an example. In the example, the series reaches a peak in period t and a trough in period t+2. The duration of the downturn is thus 2 and the amplitude ab. After hitting a trough in period t+2, the series recovers the value of b in period t+3. Thus, the time until recovery to previous peak is 1 (or t+3 minus t+2), and the growth during the first year after the trough is also ab.

3.3 Definitions of Policy Responses

A difficulty in measuring policy responses is to distinguish between discretionary changes in policies and the component of policies that automatically responds to economic fluctuations. Government spending, for example, is largely constrained by government revenue which in turn is determined by economic activity. Consequently, government spending tends to decrease during recessions even when the government switches to an expansionary stance by running deficits. In order to properly evaluate discretionary policy responses by the government, one needs to control for the automatic positive correlation between government consumption and GDP that may originate from income elasticity of tax revenue. In this paper, I address this problem by making cyclical adjustments for policy variables using a simple regression framework. More specifically, to cyclically adjust government consumption, I estimate the following equation separately for each of the developed-economy group and the developing-economy group:

government consumption gap_i,t = β₀ + β₁*GDP gap_i,t + e_i,t (1)

where government consumption gap and GDP gap are residuals from Hodrick-Prescott filtering of logarithm of government consumption expenditure and logarithm of GDP, respectively.⁷ Cyclically adjusted government consumption is given by the residual in equation (1).⁸ It can be regarded as a measure or proxy of discretionary fiscal policy, with a positive value indicating an expansionary stance.

For the call interest rate, the following equation is estimated for each of the developed-economy group and the developing-economy group:

real interest rate_i,t = γ₀ + γ₁*GDP gap_i,t + γ₂*inflation_i,t + γ₃*dummy_85_i,t + ε_i,t (2)

where dummy_85 is a dummy variable that takes the value of 1 for periods after 1985. It is included to allow for a possible structural shift in the equilibrium interest rate. A cyclically adjusted interest rate is given by the residual in equation (2). Since equation (2) corresponds to the Taylor rule for monetary policy, a positive deviation from the rule can be regarded as a discretionary tightening of monetary policy.

Now, using cyclically adjusted policy variables obtained from equations (1) and (2), I can measure the policy response over the course of a recession. Specifically, I define the policy response as the cumulative sum of changes in the cyclically adjusted policy variable during a recession period (from a peak to the next trough).

3.4 Stylized Facts

I first cite some summary statistics about recovery from Hong, Lee, and Tang (2010b). As Table 1 [ PDF 16.3KB | 1 page ] shows, the average time until recovery to the previous peak is 1.89 years for developing Asian economies while only 1.44 years for OECD economies. The slower recovery in Asia can also be seen from Figure 5 [ PDF 20.4KB | 1 page ], which illustrates the probability that an economy will remain below the previous peak beyond a certain number of years. The probability to “survive” (to remain below the previous peak) is estimated using a Weibull distribution. It is clear from Figure 5 that the survival function takes consistently higher values for developing Asia. For example, the probability to “survive” (to remain unrecovered) two years after the trough is about 50% for developing Asia and about 40% for OECD. Using the same sample of countries, Hong, Lee, and Tang (2009a) reported that both the duration and the absolute magnitude of the peak-to-trough amplitude are greater for developing Asia. Table 1 and Figure 5 indicate a recession is not only longer-lasting and more severe but also harder to overcome in developing Asia. The median length of time to recovery, however, is equally 1 year for both regions, suggesting that developing Asia has witnessed a few exceptionally slow recoveries in the past.

My second measure of recovery, the GDP growth rate during the first year of recovery, is 5.25% for developing Asia and 2.92% for the OECD countries. As mentioned above, Hong, Lee, and Tang (2009a) report that the peak-to-trough drop is also sharper in developing Asia. Taken together, these results imply that the developing Asian economies experience a steep downturn but recover quickly.. This may seem to contradict the aforementioned patterns in the time until recovery and the survival function. I note, however, that the two regions have very different potential growth rates as shown in Table 1, the all-time average of GDP growth is 3.33% for the OECD and 5.49% for developing Asia.⁹ When this fixed difference in the potential growth rate is accounted for, my second recovery measure does not show much difference between the two regions. For Korea, the first measure of recovery is one year and the second measure of recovery is 7.5%.

Next I present some findings about policy responses from Hong, Lee, and Tang (2010b). As explained above, I define policy responses as the cumulative sum of changes in cyclically adjusted government consumption or cyclically adjusted call interest rates. Thus, a positive value for fiscal policy corresponds to an expansionary stance while a positive value for monetary policy corresponds to a contractionary stance. Table 2 shows how fiscal and monetary policies have responded to recessions in OECD and developing Asia.¹⁰ I find that discretionary changes in government consumption have been significantly positive during recessions in both the OECD and developing Asia, with growth rates of 2.2% and 3.9%, respectively. Clearly, this does not necessarily mean that actual government consumption increases during a recession. With no cyclical adjustment, government consumption may well exhibit negative growth during a recession as fiscal revenues decrease. Indeed, cumulative government consumption growth during a recession turns out to be -0.5% for developing Asia and 2.7% for the OECD in my sample. Fiscal policy measures in Table 2 [ PDF 41.3KB | 1 page ] only indicate that government consumption during a recession tends to decrease less than is implied by the simple income elasticity. Also note that the difference in the fiscal policy measure between OECD and developing Asia does not necessarily imply that Asian governments have been more counter-cyclical on the whole. The fiscal policy measure may be larger in Asia because automatic stabilizers in the region are relatively small and thus need to be supplemented by discrete government spending.

In contrast to fiscal policy, the monetary policy response is significantly different from zero only in OECD economies. According to Table 2, the cumulative decrease in the cyclically-adjusted call rate during a recession is about 0.58 percentage points for OECD. And it is significantly positive at the ten % level.¹¹ For developing Asian economies, the cumulative change in the call rate is not significant even if negative, suggesting that monetary authorities in Asia may not have been as active in changing the interest rate to moderate the effects of a recession.

Korea's policy responses during the two recession episodes in 1980 and 1998 have been greater than the international average: the fiscal policy response was 11.2% and the monetary policy response was -1.55 percentage points.

3.5 Policy Responses and Recovery

In this section, I examine whether the counter-cyclical policy responses play an important role in the recovery process. To achieve that goal, I first estimate the following equation that relates GDP growth during the first post-trough year to policy responses and control variables:

GDP growth during the first year of recovery_i,t = δ₀ + δ₁*fiscal policy_i,t (3)
+ δ₂*monetary policy_i,t + δ’X_i,t + η _i,t

The unit of observation in equation (3) is each recession episode and thus the subscript i,t denotes recession t in country i. The policy measures used here are the same with those in Table 2 and the vector of control variables X includes the peak-to-trough amplitude of a recession and policy variables of Japan. The peak-to-trough amplitude is included to control for possible interrelation between the severity of a downturn and the pace of subsequent recovery. To the extent that there is overshooting in the initial drop, a greater fall will tend to be followed by a sharper recovery in the following periods. Also, since Japan is a leading economy in Asia, the macroeconomic policy stance of Japan may have important implications for the recovery of other economies in the region. Japan's policy stance is measured by the changes in cyclically adjusted interest rate and government consumption expenditures in the year immediately preceding a trough. Equation (3) is estimated using a panel model with fixed effects in order to control for cross-country differences in the potential growth rate and other country-specific effects.

In addition to equation (3), I estimate the following equation that specifies the probability density of recovery as a Weibull function:

probability density of recoveryi = f(t_i, z_i) = γe^β’Zit_i^γ-1exp(-e^β’Zit_i^γ) (4)

In my definition, a recovery is achieved when GDP, after hitting a trough, recovers to the previous peak value. Thus, the event time ti denotes the time when GDP recovers to its previous peak in the recession episode i. zi and β are the covariate vector of recession i and the corresponding coefficient vector, respectively, and γ is the shape parameter of the distribution. The covariate vector z includes the same explanatory variables as in equation (3), i.e., the peak-to-trough amplitude of a recession and policy variables of Japan. Again, the unit of observation is each recession episode rather than a country-year.

I first estimate equation (3) using GDP recessions from OECD and developing Asia and provide the results in columns (1) and (2) of Table 3 [ PDF 17.8KB | 1 page ]. I consider the two regions at the same time in order to maximize the number of observations. When considered separately, Asia has only 19 recession episodes that have no missing values in the policy and control variables. However, since the main goal of the analysis is to derive implications for Korea, I also estimate equation (3) using only the Asian sample and report the results in column (3). Fortunately, despite the small number of observations, estimation results from the Asian sample are not much different.

Column (1) shows that, for the whole sample, fiscal policy has a significantly positive effect on the pace of recovery. The estimated coefficient (about 0.3) implies that a one standard deviation increase in the fiscal policy measure (about 3.4 percentage points) leads to a 1 percentage point increase in the GDP growth rate. The IMF (2009a) has shown, using quarterly data for the OECD economies, that the fiscal policy response is important for the strength of recovery. Applying the same approach to annual data from OECD and developing Asian economies, I confirm the IMF's finding that counter-cyclical fiscal policy helps promote recovery. In contrast to fiscal policy, monetary policy does not seem to have a systematic effect on the post-trough growth: the coefficient on the monetary policy measure is close to zero and insignificant. The coefficient on the absolute value of amplitude is significantly positive, indicating that the economy tends to recover more rapidly following a greater initial fall.¹²

In column (2), I add Japan's policy variables and find that the monetary policy of Japan has a significant effect on other countries' recovery. The coefficient indicates that a 1 percentage point decrease in cyclically adjusted interest rate in Japan is associated with a 0.4 percentage point increase in the post-trough GDP growth rate in other countries. Although not reported in the table, when examining fiscal and monetary policies of the United States and Germany, I find that none of the foreign policy variables has a significant effect. This is not entirely surprising, because my sample is heavily representative of Asian economies that have strongest economic ties with Japan. However, one may still wonder exactly how a lower interest rate in Japan promotes recovery in other countries when it has not been particularly successful in providing a boost to the domestic economy. One possible explanation is that an increase in the supply of Japanese Yen may be associated with increased international capital outflows. For example, a lower interest rate in Japan encourages other countries to increase their borrowings from Japan. The increased borrowings in turn will result in an easing of financial conditions in those countries, thereby stimulating economic activity. Hong and Lee (2009) provide an example to support this argument. By applying the business cycle dating methodology of Harding and Pagan (2002) to Japan's domestic credit, they report that Japan appears to have experienced a major credit contraction between the fourth quarter of 1996 and the first quarter of 1998, just before the onset of the 1997–1998 Asian crisis. This suggests that tight credit conditions in Japan may have been a partial cause of the Asian crisis.

In column (3), I examine how the estimation results change when only Asian recessions are used in the estimation. Column (3) shows that, despite the small number of observations, coefficients on macroeconomic policy variables are still significant and of the right sign. In particular, the coefficient on fiscal policy is significantly positive and of a similar magnitude as before. Also, unlike in columns (1) and (2), monetary policy now has a significantly negative coefficient, consistent with common expectations. Overall, Table 3 supports the view that expansionary policies, particularly expansionary fiscal policies, have been effective in promoting recovery from economic recession in Asia.

I now estimate equation (4) and provide the results in Table 4. As before, columns (1) and (2) use GDP recessions from the OECD and developing Asia collectively, while column (3) considers only Asian recessions. Since equation (4) specifies the probability of recovery (or the hazard of exiting a recessionary state during which GDP remains below the previous peak), a variable with a positive coefficient should be interpreted as promoting recovery. Estimates reported in Table 4 denote coefficients rather than hazard ratios. Results in Table 4 [ PDF 34.5KB | 1 page ]are broadly consistent with those in Table 3. First, in all specifications, the coefficient on fiscal policy is significantly positive while the coefficient on monetary policy is insignificantly different from zero. Second, the coefficient on the absolute value of amplitude is always significant and negative, indicating that the larger the initial drop is the longer it takes to recover. Note that this is not necessarily inconsistent with the positive coefficient on the same variable reported in Table 3. For a recession with a large peak-to-trough drop, it is entirely possible to have a high post-trough growth and yet not to recover to the previous peak quickly. Third, when significant, an expansionary policy in Japan promotes recovery in other countries. As columns (2) and (3) show, either a decrease in Japan's interest rate or an increase in Japan's government consumption is associated with a higher probability of recovering from a recession.

Tables 3 and 4 show that an expansionary macroeconomic policy can curb negative effects of a recession in the short run. The medium-term effect of the policy, however, is not known. Considering the time lag between a policy action and its influence on the economy, an expansionary action may continue to stimulate the economy several years later. On the other hand, an expansionary policy may entail an increase in fiscal burden or inflation, which in turn may increase the possibility of a recurring recession. In order to examine the medium-term implications of an expansionary macroeconomic policy, I estimate the following probit equation that specifies the probability of a recurring recession as a function of short-term policy responses and control variables:

probability of recurrencei = Ф(α₀ + α₁*fiscal policy_i + α₂*monetary policy_i + α’X_i + η_i), (5)

where Ф is the cumulative normal distribution function and the subscript i denotes recession i. I define a recurring recession as a recession that takes place within five years of a trough. Using this definition, the probability of recurrence is 0.28 for the whole sample and 0 for Korea. I report the estimation results of equation (5) in Table 5. As before, columns (1) and (2) use the whole sample while column (3) uses only the Asian sample. One of the most robust results in Table 5 is the significantly positive coefficient on the monetary policy measure. This result indicates that a greater decrease in the interest rate made during the initial GDP downturn tends to lower the probability of another recession during the five years following the trough. Fiscal policy, on the other hand, is not particularly important in the medium term. As will be shown later, a fiscal expansion typically results in a persistent increase in fiscal debt. Table 5 suggests that, at least in the medium term, this negative side-effect of a fiscal expansion does not have a direct influence on the economy.

Another robust finding in Table 5 [ PDF 17.8KB | 1 page ] is the significantly negative coefficient on the absolute value of amplitude. This indicates that a downturn with a large amplitude (in absolute terms) is less likely to be followed by another downturn within the next five years. In other words, the economy may become more resilient to a negative shock after going through a severe recession. Policy variables of Japan are all insignificant and need no further comment.

Download this Paper [ PDF 354.1KB| 26 pages ].

Comment(s)

There are [0] comment(s) for this entry. Post a comment.