1
The Effect of Government Spending on the Length and Depth of Recessions
October 1st, 2012
Final Master Thesis, Ma Finance R.W.W. Rijnberg (s1612301)1 Supervisor: dr. J.O. Mierau Co-reader: dr. P.P.M. Smid
Keywords: Government spending, recession, fiscal policy JEL-classification: E62, H30, H50
Abstract
In this paper, both the relationships between the length or depth of recessions and the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) are tested. Both relationships are estimated by using OLS regression analyses; 47 recessions are found by using GDP growth rates of 35 countries during the period 1970-2000. Control variables are used for a country’s state (war / peace), for a country’s openness, for a country’s geographical location, and for time. No significant relationships are found between the change in government expenditures during recessions (compared to a 5-year average benchmark) and the length or depth of recessions. In the latter case, there is found a significant negative relationship between a country’s openness and the country’s recession depth, implying that more open countries experience worse recessions than relatively closed countries.
1
2
Introduction
Should governments try to trigger their economies at the start of a recession? This is a question which arises every new recession. Until now, it has been widely believed that discretionary fiscal policy is not the right method to stimulate the economy during recessions. Not, because discretionary fiscal policy doesn’t work in theory, but due to the fact that implementation takes too much time (Andolfatto, 2010). Instead, it is seen as the task of monetary authorities to smoothen the business cycle via an accommodative interest rate policy.
Since 2008, the worldwide economy has suffered from a financial crisis leading to a prolonged depression for many countries. While many people have different beliefs about the possible causes of this crisis, the bubble of asset prices (such as housing prices), the extensive and easy lending practices of banks and soft regulation have been put forward as the main reasons for why this financial crisis has started and why this crisis has been so severe (Mishkin, 2011).
This recent worldwide economic crisis has caused renewed attention to the question whether government spending as a way of stimulating aggregate economic output and employment during an economic downturn is useful (Woodford, 2011). By the end of 2008, short-term nominal interest rates had fallen to almost zero in many countries leaving monetary policy little room to stabilize economic distortions and spiraling unemployment. As a result, interests in fiscal stimulus as an option to stabilize economic downturns have greatly increased. But is fiscal stimulus indeed effective?
Current research has focused on the size of fiscal multipliers (see: Auerbach and Gorodnichenko (2012), Hall (2009), and Heim (2011)) and on the duration dependence of recessions (Castro, 2010). The first type of research deals with the effects of government spending, deferred from taxes, on national income while the second type of research tries to answer the question whether periods of expansion or contraction in economic activity are more likely to end when they become older.
In this paper, two relationships are tested. First, the relationship between the length of recessions and the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) is tested. Second, the relationship between the depth of recessions and the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) is tested. Both relationships are estimated by using OLS regression analyses. In this paper, 47 recessions are found by using GDP growth rates of 35 countries during the period 1970-2000 and these are used for regression analyses. Control variables are used for a country’s state (war / peace), for a country’s openness (in terms of trade), for a country’s geographical location (in terms of continents), and for time (in decades).
3
In the next section, a theoretical literature review will be provided. Then, an empirical literature review will be presented. After that, a data and methodology section will be given. Furthermore, the estimation results and robustness tests will be provided and a conclusion will summarize the most important findings. Finally, the appendices and references will be presented.
Theoretical Literature Review
According to Bunea-Bontas and Petre (2010), fiscal policy is the intended manipulation of government income and spending in order to achieve certain economic and social objectives and, especially, in order to sustain economic growth. It is based on the beliefs and theories of the well-known British economist John Maynard Keynes. His theories, also well-known as Keynesian economics, state that governments could affect macroeconomic productivity levels by changing tax levels and public expenditures. Changes in the composition and level of these two fiscal policy instruments affect aggregate demand and the level of economic activity, the pattern of resource allocation, and the distribution of income (Heyne, Boettke and Prychitko, 2002). A decrease in the level of taxes or an increase in government expenditures enlarges national income through rising aggregate demand; an increase in the level of taxes or a decrease in government expenditures leads to an income leakage through falling aggregate demand. So, based on the Keynesian theories, the government is able to influence the levels of national output and employment by affecting aggregate demand through fiscal policy instruments. This makes fiscal policy an important potential tool for economic stabilization. During a recession, a country might have to deal with unused productive capacity and unemployed workers. In that case, increases in demand through fiscal policy will mostly lead to more national output without changing the price level drastically; during a ‘booming’ economy, by contrast, an increase in government spending or a decrease in tax levels is most likely to impact price levels drastically while not influencing national output that much (Heyne, Boettke and Prychitko, 2002).
Keynes’ theories thus seem to suggest that an active government fiscal policy could be effective in steering the economy. Keynes advocated a countercyclical fiscal policy, which means that the government expenditures should rise (and equivalently, taxes should be cut) during times of economic downturns, i.e. deficit spending, and the opposite, i.e. raising taxes and/or decreasing expenditures, during economic booms. This would decrease unemployment rates and stabilize wages during recessions via labor-intensive infrastructure projects and suppress inflation pressures during economical periods in which there is abundant demand-side growth. Moreover, governments should act in the short run rather than waiting for market forces to solve the problems (Keynes, 1924).
4
While these neoclassical economics may be valid, Keynesian economists argue that their fiscal policies only are applicable in the case where unemployment is considerably high, i.e. above a specified non-accelerating inflation rate of unemployment. Then, they argue, the earlier mentioned ‘crowding out’ effect is minimal. Moreover, private investment could be ‘crowded in’ next to ‘crowded out’. First, fiscal stimulus by the government raises aggregate demand and hence business output, leading to a rise in cash flow and profitability which in turn enhances business optimism. This accelerator effect implies that government and private firms are complements rather than substitutes. Second, fiscal stimulus by the government leads to an increase in gross domestic product, raising the amount of saving which helps to finance the rise in fixed investment. Last, government investment in public goods, which are not provided by private firms, will stimulate the private sector’s growth. For instance, government expenditures on research, public health, education and infrastructure doubtlessly help the long-term growth of potential national output.
Keynesian economists thus think that adding to profits and income during boom cycles via tax cuts, and extracting income and profits from the economy through cuts in expenditures and/or higher taxes during recessions tends to worsen the negative effects of the business cycle. This effect is particularly distinct when the government controls a considerable part of the economy, and is therefore one reason fiscal conservatives endorse a much more passive government (Rothbard, 1947).
In order to stabilize the economy, fiscal policy could work in two ways. First, fiscal policy in itself embodies a mechanism of automatic stabilization. Secondly, discretionary fiscal policy could reduce economic instability, as well (Bunea-Bontas and Petre, 2010). Fiscal policy in itself includes automatic stabilizers which act to moderate automatically the contractions and expansions of the economy cycle. This mechanism enlarges automatically the fiscal policy during recessions and contracts it during economic booming periods; it is therefore a form of counter-cyclical fiscal policy and depends on the amount of national production and income, such that business cycle instability is automatically reduced without any need for discretionary policy. Automatic stabilizers increase government budget deficits in times of recessions and decrease them in times of economic prosperity. These stabilizers consist of income taxes and transfer payments. Both are highly dependent of national production and income and work in opposite ways. If national production and income rise, income tax collections also tend to increase, whereas transfer payments tend to fall. On the other hand, if national production and income fall, income tax collections tend to fall, whereas transfer payments tend to rise.
5
policies. Last, once a fiscal policy has been changed, it might take 6-12 months before those adjustments in fiscal policy actually affect the whole economy.
There are two types of discretionary fiscal policies. An expansionary fiscal policy involves a net increase in government expenditures; this can be achieved through a rise in government expenditures or through a fall in tax revenues or a combination of both, being associated with a budget deficit. This leads to a larger budget deficit or smaller budget surplus (Bunea-Bontas and Petre, 2010). During a recession, this type of policy could be run in order to restore national output to its normal level and to decrease unemployment rates. A contractionary fiscal policy, on the other hand, appears when net government expenditures are reduced through lower government spending, higher tax revenues or a combination of both, being associated with a budget surplus. This leads to a smaller budget deficit or a larger budget surplus. As stated before, the disadvantage of these discretionary policies is the considerable time lag between the enactment of these policies and the moment at which these policies affect the economy.
Due to time lags, persistent changes in government spending are likely to have more effect on output than temporary ones. With respect to this belief, Aiyagari, Christiano and Eichenbaum (1992) analyzed the effect on aggregate variables of changes in government consumption in a stochastic, neoclassical growth model. They showed theoretically that the effect on output and employment of a persistent change in government consumption is well above that of a temporary alteration. Furthermore, they showed that in principle there can be an analog to the Keynesian multiplier in the neoclassical growth model.
As earlier mentioned, Keynesian economists support a counter-cyclical fiscal policy, especially when unemployment rates are high. This belief was modeled theoretically by Michaillat (2012), who demonstrated why fiscal policy becomes more effective as unemployment increases in economic downturns. His theory is based on his own equilibrium unemployment model, in which jobs are allocated in recessions. Fiscal policy could take the form of government expenditures on public-sector jobs. Recessions are periods of severe job shortage without much competition for workers among hiring firms; in his model, hiring in the public sector does not crowd out hiring in the private sector much. Therefore, fiscal policy diminishes unemployment effectively. Formally, the fiscal multiplier, i.e. the decline in unemployment rate achieved by spending one dollar on public-sector jobs, is countercyclical. An implication of this statement given by the author could be that available estimates of the fiscal multiplier, which measures the average effect of fiscal policy over the business cycle, do not apply in recessions because the multiplier appears to be much higher in recessions than on average.
6
above one are possible and in that case welfare increases if government spending increases, since it fills, at least partially, the output gap arising from the inability to lower interest rates.
If it comes to nominal interest rates which are near its lower bound of zero, another important Keynesian concept that should be discussed briefly is the liquidity trap.
A liquidity trap is a circumstance in which cash injections into the private banking system by a central bank are unsuccessful to lower interest rates and thus fail to encourage economic growth. Liquidity traps often arise when a country’s residents start to hoard cash due to adverse expectations, such as deflation, insufficient aggregate demand or war. The main characteristics of a liquidity trap are short-term interest rates that are near zero and fluctuations in the monetary base that fail to translate into changes in national price levels (Gärtner and Jung, 2011).
Gärtner and Jung (2011) theoretically showed that in the presence of a liquidity trap, fiscal policy could still work, even in a small, open economy under flexible exchange rates. The authors emphasize a strong coordination between fiscal and monetary policy, however, in order to let both policies be effective in stimulating the economy.
In addition, Cook and Devereux (2011) stated that integrated financial markets tend to give rise to liquidity traps. They stated that in an international environment, fiscal policy could be effective in increasing GDP when the country faces a liquidity trap; however, it does so by ‘beggaring thy neighbor’, i.e. by having a negative cross country spillover effect of fiscal policy. Although fiscal policy coordination among countries would seem to be a good response to this problem, the authors stated that there is little case for coordinated global fiscal expansion. Countries hit by a liquidity trap should use their own policies to react without much help from foreign policies.
If a liquidity trap continues to hold for a long period of time, spending multipliers are found larger than in normal circumstances, while budgetary costs are found minimal. This would mean, provided that multipliers are larger than one, that fiscal expansion should be executed limitless. Erceg and Lindé (2010) tested this statement by examining the effects of an expansion in government expenditures if the country is in a liquidity trap. They showed that even in the case of high multipliers for small increases in government expenditures, those multipliers might decrease significantly at higher spending levels. It is thus important to distinguish between the marginal and average responses of output and government debt.
Empirical Literature Review
7
Bunea-Bontas and Petre (2010) discuss some practical problems that might affect policy decisions during an economic downturn, although fiscal policy in theory is effective. First, higher government expenditures may lead to higher interest rates, since the issue of more government debt leads to lower government bond prices and higher bond yields. This could cause a discouragement in private investments. Second, cutting taxes could lead to saving patterns of the consumer (although they have more money to spend) due to the fear of getting unemployed. Therefore, government focus should be on creating jobs. Last, in practice it is difficult to have a flexible fiscal policy; tax rates cannot be easily adjusted and government spending is to a large extent fixed in contractual terms.
If it comes to discretionary fiscal policies, it appears that advanced economies conduct different policies than emerging economies. In advanced economies, discretionary fiscal policy has been countercyclical most of the times; this means that during recessions, taxes typically have been cut and government expenditures have been increased. On the other hand, in emerging economies, discretionary fiscal policy has typically been pro-cyclical, which means that government expenditures have been decreased (or taxes increased) during economic downturns (IMF survey, 2008).
Furthermore, while Aiyagari, Christiano and Eichenbaum (1992) may have provided a theoretical ground for the influence of changes in government consumption on output, Kuester (2011) empirically estimated the effect of temporary rises in public spending on economic activity via a benchmark New Keynesian model. The author found that especially when monetary policy is constrained by the zero lower bound on interest rates, fiscal stimulus might be more effective than in less severe recessions. Moreover, the author emphasizes that fiscal policies must be carefully constructed in order to have the intended effect. It appears that fiscal stimulus is most effective in cases where the economy is still disrupted and interest rates are still near its lower bound of zero. However, higher public spending is not claimed to be a panacea for tackling the causes of why the economy got into a severe recession in the first place, nor does there seem to be a clear magnitude of the impact of public spending on the economy. Moreover, Baxter and King (1993) examined four classic fiscal policy experiments within a quantitatively restricted neoclassical model. They found that permanent alterations in government purchases can lead to short-run and long-run output multipliers that are above one. Secondly, those permanent changes in purchases have larger impacts than temporary ones. Thirdly, they found that the financing decision is quantitatively more important than the resource cost of alterations in government purchases. Last, public investment impacts private output and investment to a great extent.
8
receive. Such projects often do not create wealth in the sense that they are not productive. Instead, they appear to pull resources out of productive industries and continue the misdirection of resources that normally characterize economic downturns. So, higher levels of public expenditures do not appear to create the conditions necessary to let the economy recover. Instead, public spending crowds out private spending and productive economic activity. The author therefore concludes that governments should spend less during recessions so that firms could spend more and the private sector would be more likely to recover. These empirical findings seem to support the neoclassical view.
Proponents of fiscal stimulus argue that public expenditures are necessary to replace private investments normally lost during an economic downturn. Andolfatto (2010) stated that it is important to make a distinction between wartime crises and peacetime economic crises. What might be desirable during periods of war might not be desirable during periods of peace. Estimates of the fiscal multiplier based on wartime periods are utilized to support the proposition that a peacetime intervention could support the economy in a desirable way. By using a simple neoclassical model, the author demonstrated that the optimal fiscal policy, whether expansionary or contractionary, is independent of the size of the fiscal multiplier, while fiscal multipliers were found consistent with the wartime evidence.
Although Andolfatto (2010) did not test for the magnitude of crowd out effects, Heim (2011) was able to do so. He tested the crowd out effects of government deficits by means of adding deficit variables to consumption and investment models which widely control for other factors. Moreover, he added separate variables for deficits resulting from tax cuts and increases in expenditures. Both the effects for recession as for non-recession periods were calculated, and compared to models with average crowd out effects for the whole business cycle, and to models without any crowd out effects. Heim (2011) found that deficits appear to crowd out private consumption and investment, and those results are statistically significant, and explain substantial variance. They forecast ‘IS’ curve coefficients better than models without any crowd out effects. In both recessions and non-recessions, government spending deficits were accompanied by complete crowd out effects, leaving no apparent net stimulus effect. Moreover, for government tax cut deficits, the results were even stronger. They led to more than complete crowd out effects, creating net negative economic effects. Last, the crowd out effects were found to be approximately equal for recessions as for non-recession periods. Heim (2011) thus shows that both the amount of public spending and the components of public spending are important factors influencing the magnitude of crowd out effects.
9
Second, while Heim (2011) showed that also the amount of public spending is an important factor influencing the magnitude of crowd out effects, it might be even more important to know how large fiscal multipliers are in different circumstances. As stated, Heim (2011) found equal crowd out effects for recessions and non-recessions. Additional results were found by Auerbach and Gorodnichenko (2012), who obtained important results with respect to these fiscal multipliers. They provided three insights. First, while using regime-switching models, they found large discrepancies in the size of spending multipliers in recessions and expansions, with fiscal policy being substantially more effective in recessions than in expansions. Second, they estimated multipliers for more disaggregate spending variables which behave differently relative to aggregate fiscal policy shocks, with military spending having the largest multiplier. Third, these authors showed that controlling for predictable parts of fiscal shocks tends to increase the magnitude of the multipliers in recessions. Moreover, Hall (2009) gave an overview of output multipliers of government expenditures within different types of models. He showed that during World War II and the Korean War, real GDP growth was approximately half the rise in government purchases. With other factors holding back GDP growth during those periods of war, the output multipliers of government expenditures may be in the range of 0.7 to one according to the author. This range is widely supported by research based on vector auto regressions that control for other determinants, although higher values of these multipliers are not excluded. Moreover, New Keynesian models yield multipliers in that same range, as well. However, neoclassical models produce in general much lower multipliers, since they forecast falling consumption whenever government purchases increase. Last, output multipliers may be higher (around 1.7) when the nominal interest rate is at its lower bound of zero.
If it comes to the transmission of fiscal policy shocks into economic activity, the confidence of households and businesses seems to play an important role. Bachmann and Sims (2011) examined this assumption by using standard structural VARs with government expenditures and aggregate output augmented to include empirical measures of consumer or business confidence. Furthermore, they also estimated non-linear VAR specifications to allow for differential impacts of government expenditures in ‘normal’ times versus economic downturns. In normal times, confidence does not appear to respond significantly to unexpected rises in government spending and spending multipliers are approximately equal to one. During recessions, confidence improves and spending multipliers are significantly larger. The authors then measure the importance of the systematic response of confidence to spending shocks for the spending multiplier and found that, in normal times, confidence appears to be irrelevant for the transmission of government spending shocks to output, but during periods of economic downturn, confidence appears to be significantly important. It is not confidence per se, in the sense of pure sentiment, which matters for the transmission of spending shocks during downturns, but rather the composition of expenditures during a downtown is different. More precisely, spending shocks during recessions forecast future productivity improvements through a persistent rise in government investment relative to consumption, which is in turn reflected in higher measured confidence.
10
government budget balance depend on the relative size of the marginal propensity to consume and invest and could be positive under certain conditions. Their empirical estimates showed that consumption and production structure have changed considerably from 1930 to 2007; both positive and negative effects on growth and budget balance of the same fiscal policy were found in different time periods.
Recent developments of near-zero nominal interest rates in Japan and the US have raised new questions about the conduct of fiscal policy. Christiano, Eichenbaum and Rebelo (2009) found that fiscal policy is especially effective when the nominal interest rate is zero in a New Keynesian model. In their paper, these authors examined a 5 percent shock to the preference discount rate that results in a binding zero lower bound on the nominal interest rate. They found an output multiplier with respect to a rise in government purchases of about four, which is substantially large. Through solving a log linearized version of the model that is centered around a steady zero inflation rate, this result about the dynamics of the New Keynesian model with a zero nominal interest rate was found statistically significant. More researchers, such as Eggertsson and Woodford (2003), Christiano (2004), Braun and Waki (2006), and Eggertsson (2009) used log linearized solutions to analyze the different properties of New Keynesian models if nominal interest rates are zero. The research of Christiano, Eichenbaum and Rebelo (2009) thus indicated that there seems to be an important role for the fiscal authority in stabilizing the economy when monetary policy is constrained by the zero lower bound on the nominal interest rate. However, Braun and Waki (2010) found that log linear methods produce large biases that misrepresent the magnitude of the government purchase output multiplier by a factor of two. This upward bias associated with the log linear solution is due to two factors, namely a factor related to the resource costs of price adjustment or dispersion and secondly, these resource costs have an alleviating effect on the output response to a preference discount factor shock and this enforces to weaken the response of marginal costs.
The last field of empirical literature that should be discussed here is the duration dependence of recessions. Although a recession’s duration dependence, i.e. the likelihood of a recession ending depending on its age, is not subject to this paper, this concept has gotten increased attention the last two decades and should be discussed in order to provide a more complete view of the empirical literature concerning recessions and fiscal policy.
Various papers have tried to answer the question whether periods of expansion or contraction in economic activity are more likely to end when they become older. To answer this question, several models have been put forward, such as parametric, non-parametric duration models, and Markov-switching models (Castro, 2010).
First, non-parametric duration models have not provided significant results overall if it comes to finding proof of duration dependence for economic expansions and recessions.
11
expansions in France, Germany and the UK. Last, Abderrezak (1998) also used a parametric hazard model to examine duration dependence in eleven industrial countries. However, instead of using the classical business cycles, the author used growth cycles. Both individual country regression as pooled regression gave significant results with respect to positive duration dependence for both growth phases as the whole growth cycles.
Third, Markov-switching models have been used to examine positive duration dependence for economic expansions and contractions. For instance, Durland and McCurdy (1994) provided proof of duration dependence for recessions in the US after WWII by using a Markov-switching model. However, no evidence of duration dependence was found for US expansions after WWII. Lam (2004) extended this research by means of allowing for heteroscedasticity in the noise component and through allowing for duration dependence not only in transition probabilities but also in mean growth rates. The author found that the probability of an expansion ending decreases gradually as the expansion ages, whereas the probability of contractions ending increases significantly when the contraction ages.
Last, other types of econometric models have been used to examine duration dependence for business cycles. Both Di Venuto and Layton (2005) and Layton and Smith (2007) established a multinomial regime-switching logit model to test for duration dependence in Australian and US business cycles, respectively. Both found positive duration dependence for both expansions and contractions.
One of the main limitations in these papers so far, is that little attention has been given to the possible effects of other factors. Duration dependence might be found present, but other underlying mechanisms could also influence the likelihood of an expansion or recession ending apart from a business cycle’s length.
Castro (2010) therefore utilized a much richer class of conditioning variables. More specifically, while testing for positive duration dependence for business cycles, he controlled for the effect of a composite leading indicator; furthermore, he controlled for the effects of some of the composite leading indicator’s components and for the effects of some other potential explanatory factors. In this paper, Castro used a discrete-time duration model with a panel of thirteen industrial countries over the period 1948-2006. Castro found significant evidence of positive duration dependence for both expansions and contractions. However, he also found that the duration of expansions or contractions is not only dependent on their age. More precisely, the duration of expansions appears to be also positively dependent on the behavior of the variables in the composite leading indicator and on private investment, while it appears to be negatively influenced by the price of oil and by the occurrence of a peak in the US business cycle. On the other hand, the duration of a contraction appears to be negatively influenced by its actual age and by the duration of the previous expansion. Last, no support was found for the hypothesis that fiscal rules caused European recessions to be lengthened, nor any support was found for the influence of political factors on the duration of business cycles.
12
expenditures leads to economic growth under normal circumstances, i.e. when interest rates are not near its lower bound. Instead, it appears to lead to less economic growth.
In this paper, therefore, these findings will be followed. Government expenditures are assumed to crowd out private investment. The more a government spends, the worse will be the effect on the economy. The higher the extra government expenditures during a recession, the longer this recession therefore will continue to hold. For that reason, it is also assumed that cutting government expenses during a recession leads to shorter recessions. Hence, the first hypothesis is:
H1: There is a positive relationship between the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) and the length of recessions.
Since earlier empirical findings found that extra government expenditures lead to less economic growth, extra government spending during recessions is assumed to worsen the recessions. That is, the depth of recessions, measured by the negative growth rate, is assumed to increase, i.e. to become more negative, the more is spent by national governments. For that reason, it is also assumed that cutting government expenses during a recession therefore leads to less ‘deep’ recessions. Thus, the second hypothesis is:
H2: There is a negative relationship between the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) and the depth of recessions.
Data
In this paper, the first dependent variable for which OLS regressions are estimated is the length of recessions. The length of recessions is measured in yearly quarters, with a minimum of 2 quarters. The recession is assumed to start whenever a country faces two subsequent quarterly negative growth rates in Gross Domestic Product (GDP) and ends whenever this negative growth turns into positive numbers (Shiskin, 1974).
The second dependent variable for which OLS regressions are estimated is the depth of recessions. The depth of recessions is measured as the average negative growth per yearly quarter during a recession, measured in percentages.
In this paper, the independent variable for both the length and depth of recessions is the change in average government expenditures per GDP during the recession compared to the 5-year average government expenditures per GDP before the recession, measured in percentage points per yearly quarter.
13
more dependent on other countries and hence the effect of government spending on the length or depth of recessions is assumed to be less pronounced.
Second, a country’s state (war / peace) is controlled for by using a dummy variable (1 = war; 0 otherwise). Barro and King (1984) estimated the impact of temporary, wartime government purchases on output and they found statistically significant multipliers of 0.85 (formal estimate) and 0.60 (informal estimate). These findings are significantly different from the multipliers found in non-war periods (Auerbach and Gorodnichenko, 2012). Therefore, a country’s state must be controlled for in the regression analyses.
Last, both time effects and geographic effects are controlled for by using slope and intercept dummy variables for each continent and decade.2 It should be noted that Mexico is included in the South America dummy variable; furthermore, New Zealand is included in the Australia dummy variable, and Turkey is included in the Europe dummy variable. A detailed description of the variables is available in Table 1. below:
Table 1.
Definition of the regression variables
Variables Definition
RECESSIONLENGTHt = the recession’s length, measured in yearly quarters;
RECESSIONDEPTHt = the recession’s depth, measured by the recession’s average (negative) GDP
growth per yearly quarter;
∆GOVEXPt = the change in average government expenditures per GDP during the recession
compared to the 5-year average government expenditures per GDP before the recession, measured in percentage points per yearly quarter;
DNAt = 1 if the recession took place in North America, 0 otherwise; DSAt = 1 if the recession took place in South America, 0 otherwise; DEURt = 1 if the recession took place in Europe, 0 otherwise;
DASIAt = 1 if the recession took place in Asia, 0 otherwise;
DAFt = 1 if the recession took place in Africa, 0 otherwise;
D1970-1980t = 1 if the recession took place in the time period 1970-1980, 0 otherwise;
D1980-1990t = 1 if the recession took place in the time period 1980-1990, 0 otherwise;
OPENNESSt = the average total of imports and exports per GDP during the recession as a
percentage per yearly quarter;
DWARt = 1 if a country is in war during a recession period, 0 otherwise;
ε = error term.
Notes:
◦ The continent left out of the regressions is Australia. Furthermore, the time period 1990-2000 was excluded from
the regression analyses; these continent and time dummies are excluded from the regression analyses in order to avoid perfect multicollinearity.
2
The data on total GDP and GDP growth
The GDP growth rates were used to indicate the recessions. are OECD countries) are used for the period 197
appendix A. In total, 47 recessions a
distribution of recessions per decade can be found in Figure 1 First, as Figure 1. points out, most recessions
Figure 1. shows the relative distribution of recessions per decade, the sum of distributions must add to one for each decade
period 1970-1980, relative distributio
Last, there are found six recessions in the period 1970 1980-1990, and 29 in the period 1990
of recessions over time could unfortunately GDP data of the period before 19
in the number of recessions for the period 1990 available for that period. For that reason,
quarterly data is given per decade in Figure 2
from before the nineties, has caused the sharp increase in the absolute number of recessions from twelve in 1980-1990 to 29 in 1990
total available data is around eleven
however, there are still twice as much recessions in both the eighties and the nineties as in the seventies. This does not mean, however, that countries, such as the Netherlands or Germany, started to experience twice as much recessions
seventies, although at first sight this statement seems valid. However, after having taken a closer look to the data used in this paper, it appears that especially ‘weaker’ countries, i.e. countries having a tendency of experiencing recessions more often than stronger countries, do not provide GDP data for the whole period. As a result, the relative number of recessions rose in the eighties due to the inclusion of data from the weaker countries
0.00% 7.50% 15.00% 22.50% 30.00% 37.50% 1970-1980 R el a ti v e d is tr ib u ti o n o f re ce ss io n s p er d ec a d e (%)
Figure 1. : Relative distribution of recessions per decade
14
GDP and GDP growth rates are retrieved via the Worldbank and OECD.S The GDP growth rates were used to indicate the recessions. In total, 35 countries (of which most
used for the period 1970-2000. A list of these countries is provided in In total, 47 recessions are found and used for regression analyse
decade can be found in Figure 1. A few remarks have to be made. most recessions lasted between two and five quarters.
the relative distribution of recessions per decade, the sum of
must add to one for each decade. Third, since there are only six recessions for the 1980, relative distributions of recessions are either zero or high (at least 16.67 %
Last, there are found six recessions in the period 1970-1980, twelve recessions in the period 1990, and 29 in the period 1990-2000. The fact that there is a large increase in the n
unfortunately be due to the availability of data. For many countries, GDP data of the period before 1985 is not available. This fact might have caused
in the number of recessions for the period 1990-2000, simply because there is more data For that reason, the fraction of recession quarters to total available quarterly data is given per decade in Figure 2 (in %). Clearly, data availability, or the lack of data as caused the sharp increase in the absolute number of recessions from 1990 to 29 in 1990-2000, since the relative number of recessions with respect to total available data is around eleven-twelve percent for both periods. Even in relative t however, there are still twice as much recessions in both the eighties and the nineties as in the
This does not mean, however, that countries, such as the Netherlands or Germany, started to experience twice as much recessions while entering the eighties with respect to the , although at first sight this statement seems valid. However, after having taken a closer look to the data used in this paper, it appears that especially ‘weaker’ countries, i.e. countries cy of experiencing recessions more often than stronger countries, do not provide or the whole period. As a result, the relative number of recessions rose in the eighties due to the inclusion of data from the weaker countries.
1980 1980-1990 1990-2000
Time (decades)
: Relative distribution of recessions per decade
ieved via the Worldbank and OECD.Stat. In total, 35 countries (of which most A list of these countries is provided in used for regression analyses. The relative remarks have to be made. e quarters. Second, since the relative distribution of recessions per decade, the sum of these relative , since there are only six recessions for the
or high (at least 16.67 %).
1980, twelve recessions in the period 2000. The fact that there is a large increase in the number due to the availability of data. For many countries, 85 is not available. This fact might have caused a sharp increase because there is more data the fraction of recession quarters to total available Clearly, data availability, or the lack of data as caused the sharp increase in the absolute number of recessions from 2000, since the relative number of recessions with respect to Even in relative terms, however, there are still twice as much recessions in both the eighties and the nineties as in the This does not mean, however, that countries, such as the Netherlands or Germany, entering the eighties with respect to the , although at first sight this statement seems valid. However, after having taken a closer look to the data used in this paper, it appears that especially ‘weaker’ countries, i.e. countries cy of experiencing recessions more often than stronger countries, do not provide or the whole period. As a result, the relative number of recessions rose in the eighties
: Relative distribution of recessions per decade
15
Since the current crisis is still going on, it might not be a wise decision to include current data, because the length and depth of current recessions cannot be determined yet. Moreover, current GDP data are often revised a few years later. Last, many data were missing for the period 2000-2012, making proper regression analyses impossible for that period. Therefore, in order to get the correct data, the period 1970-2000 is used. Trade data were taken from the Penn World Table version 7.0 in order to construct the control variable of openness. The lists of wars (inter-state, intra-state and extra-state) were taken from Sarkees and Wayman (2010). An overview of descriptive statistics for the regression variables is given in Table 2.
Table 2.
Descriptive statistics for the regression variables
Variables Mean Median Min Max σ Observations
RECESSIONLENGTH 4.021 4 2 10 1.950 47 RECESSIONDEPTH -1.911 -1.188 -11.738 -0.072 1.986 47 ∆GOVEXP -1.721 0.195 -22.858 12.939 7.602 47 DNA* ∆GOVEXP -0.430 0.000 -20.126 12.939 3.845 47 DSA* ∆GOVEXP -0.458 0.000 -22.858 0.828 3.341 47 DEUR* ∆GOVEXP 0.291 0.000 -17.508 8.854 3.452 47 DASIA* ∆GOVEXP -0.682 0.000 -19.559 6.162 3.615 47 DAF* ∆GOVEXP -0.043 0.000 -6.616 2.021 1.057 47 D1970-1980* ∆GOVEXP 0.424 0.000 -5.298 12.939 2.570 47 D1980-1990* ∆GOVEXP -0.915 0.000 -20.126 6.162 4.371 47 OPENNESS 9.946 9.681 1.115 23.326 5.903 47 DWAR 0.234 0.000 0.000 1.000 0.428 47 Notes:
◦ Please refer to Table 1. for variable definitions;
◦◦ The sample consists of 47 recessions during the period 1970-2000.
0.00% 2.00% 4.00% 6.00% 8.00% 10.00% 12.00% 14.00% 1970-1980 1980-1990 1990-2000 R ec es si o n q u a rt er s w .r .t . to ta l a v a il a b le q u a rt er ly d a ta p er d ec a d e( %) Time (decades)
16
Except for the variable OPENNESS, all other variables are either significantly positively or negatively skewed3; hence, median values are more representative. The median recession in this sample lasts a year, and has a median quarterly growth rate of about -1.2 %. While the change in average government expenditures per GDP during the recession compared to the 5-year average government expenditures per GDP before the recession differs substantially per recession, the continent Africa shows the least deviation in this variable compared to the other continents. Furthermore, the time period 1970-1980 shows the least deviation in this variable compared to the other time periods. One should be cautious, however, if it comes to drawing proper conclusions. The conclusion that countries in Africa and countries experiencing recessions in the seventies have been pursuing more common policies (and hence less deviation in ∆GOVEXP) with respect to their national government expenditures compared to the other continents and time periods is largely influenced by the fact that Africa only includes one country experiencing recessions, i.e. South Africa, in this sample and the time period 1970-1980 only includes six recessions in this sample.
Two comments should be noted. First, since the dataset used in this paper only covers the period 1970-2000 and given the fact that Argentina was still facing a recession during the fourth quarter of 2000, the recession might have actually been longer. This idea is confirmed by Weisbrot and Sandoval (2007), who described that the Argentine economic crisis lasted till 2002. The length of this recession in this dataset is therefore understated.
Second, with respect to the recessions found in South Africa, Byrnes (1996) stated that only after 1994 national accounts were improved by reorganizing economic and legal data through the inclusion of data from all citizens of all races. The data used in this paper for South Africa from before 1994 might therefore be incomplete.
Methodology
In this paper, two relationships are tested. First, the relationship between the length of recessions and the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) is modeled by using Ordinary Least Squares (OLS) regression analysis. Second, the relationship between the depth of recessions and the change in average government expenditures per GDP during a recession (compared to a five-year average benchmark before the recession) is tested by using OLS regression analysis, as well: (1) where 4 ≤ ≤ 10, 1 ≤ ≤ 7, 11≤ ! ≤ 17,
and where in the first part of the paper represents the recession’s length, measured in yearly quarters. In the second part of the paper, represents the recession’s depth, measured by the recession’s average (negative) GDP growth.
3
17
Furthermore, the variable ∆GOVEXPt represents the change in average government expenditures per GDP during the recession compared to the 5-year average government expenditures per GDP before the recession, measured in percentage points per yearly quarter. That is:
"
#$%&'(*%$()%% +,
-
#(* ./() ,(2) where $ is the number of yearly quarters for a recession year t in which there is at least partly a recession. So, if the recession starts in Q3, $ is 2 for that year. Since not all recessions start at Q1 and end at Q4, a weighted average is used. The fraction 0123
14536 represents the relative government expenditures with respect to GDP for a recession year t. And i represents the years
tr-1, tr-2, tr-3, tr-4, and tr-5 where tr is the year in which the recession starts. If another recession
takes place within the years tr-1 – tr-5, this period will not be included in the 5-year average.
Instead, an earlier year will be added to the average. So, if a recession takes place in tr-3, tr-6 will
be used instead of tr-3, provided that tr-6 is a recession-free year. So, if two recessions for a
country took place in 1970 and 1975, the 5-year average of the ‘before recession period’ of the first recession will consist of the years 1965-1969, while for the second recession it will consist of the years 1969, 1971-1974, since 1970 is a recession year.
The variables and represent two control variables, where the first covers a country’s state (war / peace) and the second a country’s openness.
In order to construct the war dummy variable , a list of wars for the period 1970-2000 was needed for the 35 countries involved in this paper. This list was provided by Sarkees and Wayman (2010). These authors made a distinction between four types of wars, namely extra-state, inter-extra-state, intra-state and non-state wars. Extra-state wars are wars that take place between a state(s) and a non-state entity outside the borders of the state, whereas inter-state wars are wars that take place between or among states, i.e. members of the interstate system. Moreover, intra-state wars are wars that predominantly take place within the recognized territory of a intra-state, whereas non-state wars are wars between or among non-state entities, such as non-territorial entities or non-state armed groups. In order to construct their list of wars, Sarkees and Wayman (2010) used 5 rules in order to classify a period as a war. First, the war must involve sustained combat. Second, there is a presence of organized armed forces. Third, as a result, there should be at least a thousand battle-related combatant fatalities within a year period. Furthermore, the armed forces on both sides must be capable of effective resistance. This rule leads to exclusion of massacres, one-sided state killings, or riots by unorganized individuals. Last, for a state to be recognized as a war participant, the minimum rule is that it should send at least a thousand troops to the war or suffer at least a hundred battle-related deaths.
18
The variable is measured by the average total of imports and exports per GDP during the recession as a percentage per yearly quarter. That is:
#$%&'#)789,%(*% +,
$% (3)
where, once again, $ is the number of yearly quarters for a recession year t in which there is at least partly a recession. This average is weighted. The fraction 0#2:;<=,3
1453 6 represents the total of a country’s imports and exports together relative to GDP for a year t in which there is at least partly a recession.
The dummy variables, , s.t. 1 ≤ ≤ 7, control for geographic and time effects.4 Each dummy variable represents a continent or time period (in decades). For instance, represents the dummy variable for North America. Accordingly, data of countries of North America were assigned a ‘1’ and ‘0’ otherwise. The dummy variable, >, for instance, represents the time period 1970-1980, and accordingly, data of that period were assigned a ‘1’ and ‘0’ otherwise. In order to avoid perfect multicollinearity, a problem also known as the dummy variable trap, one continent and one time period were not included as a dummy variable into the regressions (Brooks, 2008).5
Last, the intercept of the regression is given by . The error term is represented by ε.
In this article, in order to define recession periods for the 35 countries involved, the most applied definition of recession is used, which is two down consecutive quarters of GDP (Shiskin, 1974). There are found 47 recession periods for the 35 countries involved for the period 1970-2000. Equation (1) is estimated via the statistical software program EViews. Equations (2) and (3) are calculated through Excel.
Finally, several statistical tests are performed in Eviews to check the assumptions of the Classical Linear Regression Model (CLRM). First, both the White and ARCH tests are done to test for heteroscedasticity among the residual errors. Then, a Durbin-Watson is performed to test for autocorrelation among the residual errors. Moreover, a Jarque-Bera test is executed to find out whether the residual errors are normally distributed. Last, a Ramsey Reset Test is executed to see whether the correct form of the regression is indeed linear as expected. All together, these tests will provide a basis check for statistical outcomes.
Results
Estimation results
Equation (1) is estimated via the statistical software program EViews by using Ordinary Least Squares (OLS). The results of the recession length regression model can be found in Table 3.
4
Please refer to Table 1. for variable definitions. In the results, the names of continents and periods will be used instead of using the numbers 1 till 7.
5
19
This table provides eight estimations; the difference between these eight estimations lies within the inclusion or exclusion of control variables for war, openness, time and geographic location. Furthermore, Table 3 includes the test statistics of these estimations. The results of the recession depth regression model can be found in Table 4. This table also provides eight estimations; the difference between these eight estimations lies, once again, within the inclusion or exclusion of control variables for war, openness, time and geographic location. Furthermore, Table 4 includes the test statistics of the recession depth regression estimations.
For the recession length regression model, the intercept (β0) is significant at a 1% level for all
eight estimations. At a 5% level, only the intercept dummy for Africa in columns (5) and (7) is statistically significant and has a positive sign; for columns (6) and (8) the same holds at a 10% level. The intercept dummy for South America in column (7) is also statistically significant at a 10% level and has a positive sign. Moreover, the slope dummy for South America in columns (7) and (8) is statistically significant at a 10% level and also has a positive sign. All other control variables did not provide any statistically significant results and the test statistics show no proof of heteroscedasticity nor of autocorrelation among the residual errors. The linear model seems appropriate according to the Ramsey Reset Test. However, for all eight estimations, (adjusted) R2 is found low6 and there’s significant proof of non-normality (at a 1% level).
Most importantly, no significant relationship is found between the change in government expenditures during recessions (compared to a 5-year average benchmark) and the length of recessions. Therefore, the hypothesis that there is a positive relationship between a change in government expenditures during recessions and the length of recessions is not supported by the data used in this paper.
For the recession depth regression model, the intercept (β0) is found significant at a 1% level for
columns (1) and (2). At a 1% level, only the intercept dummy for Asia in columns (5) and (7) is found statistically significant and has a negative sign; for columns (6) and (8) the same holds at a 5% level. The intercept dummy for South America in columns (6), (7) and (8) is also statistically significant at a 10% level and has a negative sign. The war dummy variable in column (6) is found significant at a 10% level and has a negative sign, as well. Countries in war appear to experience more severe recessions in terms of negative growth compared to countries in peace, which adds to the findings of low wartime government multipliers by Barro and King (1984). Moreover, the control variable OPENNESS is found significant overall at a 5% level and has a negative sign. The more open a country is in terms of trade with respect to national output, the deeper the recession faced by a country is. All other control variables did not provide any statistically significant results and the test statistics show no proof of autocorrelation among the residual errors. However, in columns (3) till (7), there is proof of heteroscedasticity among the residual errors at a 5% level. Moreover, the linear model doesn’t seem to be appropriate according to the Ramsey Reset Test for the regressions in columns (3), (4), (6) and (8) at a significance level of 10% and three times 1%, respectively. Furthermore, for all eight estimations, (adjusted) R2 is found low and there’s significant proof of non-normality overall (at a 1% level).
6
20
Most importantly, no significant relationship is found between the change in government expenditures during recessions (compared to a 5-year average benchmark) and the depth of recessions. Therefore, the hypothesis that there is a negative relationship between a change in government expenditures during recessions and the depth of recessions is not supported by the data used in this paper.
Overall, there appears to be no significant relationship between the change in government expenditures during a recession (compared to a 5-year average benchmark) and a recession’s length or depth.
An important note here is that the Classical Linear Regression Model (CLRM) assumes that the covariance between the error terms over time or cross-sectionally is zero (Brooks, 2008). So, it is assumed that the errors are uncorrelated with one another. In this paper, it is very likely that recessions within the same country are correlated, causing the covariance between the residual errors to deviate from zero. As a result, the coefficient estimates derived using OLS could be inefficient, leading to wrong standard error estimates (Brooks, 2008). Future research might deal with this problem, by performing regression analyses for each country separately. This is only possible, however, if the data availability per country increases and the time span used is increased, as well. Only in that case, enough recessions per country are found in order to do proper research.
21 Table 3.
Regression results and test statistics for the length of recessions
?@A@BBCDEF@EGHI JK JLMGDN@OPQ JRSTU?V JWDP@EE@BBV X JYZ[V X J\Z[VMGDN@OPQ ] where the time and continent dummy variables are represented by Djt and 4 ≤ Y ≤ 10, 1 ≤ [ ≤ 7, 11≤ \ ≤ 17
22 Table 4.
Regression results and test statistics for the depth of recessions
?@A@BBCDES@PHI JK JLMGDN@OPQ JRSTU?V JWDP@EE@BBV X JYZ[V X J\Z[VMGDN@OPQ ] where the time and continent dummy variables are represented by Djt and 4 ≤ Y ≤ 10, 1 ≤ [ ≤ 7, 11≤ \ ≤ 17
23 Robustness Tests
In order to check the robustness of the estimation results, the same regression models are performed on a dataset consisting of only OECD countries. In this way, the recessions of weaker7 countries, such as South Africa and Argentina, are not included. Moreover, the problem of incomplete or understated data, as described in the data section, is avoided in this way. In total, 39 recessions are found during the period 1970-2000. A list of OECD countries used in this paper can be found in appendix A.
The regression results are found in Table 5 and Table 6. For the OECD recession length regression model, the intercept (β0) is significant at a 1% level for all eight estimations. All
(other) variables did not provide any statistically significant results and the test statistics show no proof of heteroscedasticity nor of autocorrelation among the residual errors. The linear model seems appropriate according to the Ramsey Reset Test. However, for all eight estimations, (adjusted) R2 is found low and there’s significant proof of non-normality (at a 1% level). Most importantly, no significant relationship is found between the change in government expenditures during recessions (compared to a 5-year average benchmark) and the length of recessions.
Compared to the outcomes based on the original dataset, the main results have not changed. In fact, only the intercept dummy for South America in column (7) of the original outcomes, which is found statistically significant at a 10% level and has a positive sign, is now found statistically insignificant.8 So, the estimation results of the original dataset are found robust.
For the OECD recession depth regression model, the intercept (β0) is found significant in all
columns (mostly at a 5% level), except for columns (6) and (8). At a 1% level, only the intercept dummy for South America is found statistically significant and has a negative sign; The intercept dummy for Asia in column (5) is also statistically significant at a 10% level and has a negative sign. All other variables did not provide any statistically significant results. The test statistics show no proof of autocorrelation among the residual errors. Moreover, there is no proof of heteroscedasticity among the residual errors. Also, the linear model doesn’t seem to be appropriate according to the Ramsey Reset Test for the regression in column (6) at a 1% level. Furthermore, for all eight estimations, (adjusted) R2 is found low and there’s significant proof of non-normality overall (mostly at a 1% level). Most importantly, no significant relationship is found between the change in government expenditures during recessions (compared to a 5-year average benchmark) and the depth of recessions.
Compared to the outcomes based on the original dataset, the main results have not changed. Only, the control variables OPENNESS and DWAR are found statistically insignificant, while they are found significant in the original dataset. So, overall, the estimation results of the original dataset are found robust.
7
i.e. countries experiencing more often and deeper recessions than other countries. 8
24 Table 5.
Regression results and test statistics for the length of recessions for OECD countries
?@A@BBCDEF@EGHI JK JLMGDN@OPQ JRSTU?V JWDP@EE@BBV X JYZ[V X J\Z[VMGDN@OPQ ] where the time and continent dummy variables are represented by Djt and 4 ≤ Y ≤ 10, 1 ≤ [ ≤ 7, 11≤ \ ≤ 17
Variables / Tests (1) (2) (3) (4) (5) (6) (7) (8) β0 3.748 [0.000]*** 3.647 [0.000]*** 4.064 [0.000]*** 3.934 [0.000]*** 3.254 [0.000]*** 3.406 [0.001]*** 3.119 [0.001]*** 3.373 [0.002]*** ∆GOVEXP 0.003 [0.934] 0.001 [0.981] 0.005 [0.885] 0.003 [0.932] -0.001 [0.987] 0.001 [0.981] -0.128 [0.243] -0.138 [0.233] DWAR 0.479 [0.480] 0.423 [0.539] 0.055 [0.945] 0.327 [0.728] OPENNESS -0.032 [0.495] -0.028 [0.557] -0.020 [0.707] -0.032 [0.612] DNA 1.529 [0.143] 1.436 [0.211] 1.561 [0.243] 1.218 [0.426] DSA 0.746 [0.693] 0.888 [0.655] 0.933 [0.644] 1.149 [0.591] DEUR 0.916 [0.266] 0.943 [0.291] 1.125 [0.234] 1.069 [0.313] DASIA 0.138 [0.904] 0.211 [0.859] 0.420 [0.791] 0.357 [0.829] D1970-1980 -0.951 [0.256] -0.933 [0.281] -1.408 [0.145] -1.411 [0.166] D1980-1990 -0.587 [0.396] -0.493 [0.532] -0.691 [0.397] -0.396 [0.680] DNA*∆GOVEXP 0.111 [0.484] 0.100 [0.543] DEUR*∆GOVEXP 0.084 [0.580] 0.081 [0.607] DASIA*∆GOVEXP 0.151 [0.308] 0.152 [0.325] D1970-1980* ∆GOVEXP 0.139 [0.426] 0.171 [0.373] D1980-1990* ∆GOVEXP 0.032 [0.803] 0.061 [0.671] R² 0.000 0.014 0.013 0.024 0.115 0.119 0.187 0.201 Adjusted R² -0.027 -0.041 -0.042 -0.060 -0.085 -0.154 -0.188 -0.265 Skewness (SE) 1.519 1.569 1.594 1.640 1.427 1.479 1.525 1.661 Kurtosis (SE) 6.593 6.943 7.009 7.298 6.317 6.612 6.622 7.303 Jarque-Bera (SE) 35.989 [0.000]*** 41.264 [0.000]*** 42.644 [0.000]*** 47.509 [0.000]*** 31.122 [0.000]*** 35.431 [0.000]*** 36.424 [0.000]*** 48.020 [0.000]*** White (F-stat.) 0.109 [0.897] 0.164 [0.955] 0.262 [0.931] 0.083 [0.969] 0.351 [0.923] 0.314 [0.964] 0.313 [0.981] 0.242 [0.996] Arch (F-stat.) 0.536 [0.843] 0.621 [0.777] 0.496 [0.871] 0.562 [0.823] 0.583 [0.808] 0.548 [0.834] 0.442 [0.906] 0.470 [0.888] Durbin-Watson 1.833 1.852 1.941 1.940 1.971 2.022 1.937 1.988 Ramsey Reset quadratic (F-stat.) 0.552 [0.462] 0.108 [0.744] 0.579 [0.452] 0.454 [0.505] 0.048 [0.828] 0.173 [0.680] 0.024 [0.878] 0.002 [0.961] Notes:
◦ Please refer to Table 1. for variable definitions. The corresponding p-value is provided between brackets; *
significance at < 0.10 level; ** significance at < 0.05 level; *** significance at < 0.01 level.
◦◦The sample for OECD countries consists of 39 recessions.
◦◦◦ In order to avoid perfect multicollinearity, the intercept and slope dummy variables for Australia, Africa and the
25 Table 6.
Regression results and test statistics for the depth of recessions for OECD countries
?@A@BBCDES@PHI JK JLMGDN@OPQ JRSTU?V JWDP@EE@BBV X JYZ[V X J\Z[VMGDN@OPQ ] where the time and continent dummy variables are represented by Djt and 4 ≤ Y ≤ 10, 1 ≤ [ ≤ 7, 11≤ \ ≤ 17
Variables / Tests (1) (2) (3) (4) (5) (6) (7) (8) β0 -1.684 [0.000]*** -1.732 [0.000]*** -1.216 [0.010]** -1.257 [0.014]** -1.203 [0.033]** -1.011 [0.118] -1.170 [0.075]* -1.058 [0.167] ∆GOVEXP 0.005 [0.873] 0.004 [0.901] 0.008 [0.786] 0.008 [0.807] -0.003 [0.911] -0.002 [0.956] 0.062 [0.447] 0.056 [0.510] DWAR 0.226 [0.696] 0.134 [0.817] -0.061 [0.918] -0.101 [0.886] OPENNESS -0.047 [0.232] -0.046 [0.255] -0.025 [0.526] -0.015 [0.743] DNA -0.256 [0.731] -0.313 [0.703] -0.247 [0.801] -0.232 [0.838] DSA -4.986 [0.001]*** -4.818 [0.002]*** -5.045 [0.002]*** -4.930 [0.005]*** DEUR -0.256 [0.666] -0.184 [0.775] -0.337 [0.627] -0.261 [0.739] DASIA -1.470 [0.082]* -1.378 [0.117] -1.526 [0.201] -1.491 [0.233] D1970-1980 -0.071 [0.906] -0.046 [0.940] 0.019 [0.978] 0.064 [0.932] D1980-1990 -0.075 [0.881] -0.004 [0.995] -0.059 [0.922] -0.018 [0.980] DNA*∆GOVEXP -0.016 [0.890] -0.019 [0.879] DEUR*∆GOVEXP -0.013 [0.909] -0.011 [0.926] DASIA*∆GOVEXP -0.081 [0.460] -0.076 [0.508] D1970-1980* ∆GOVEXP -0.068 [0.598] -0.070 [0.623] D1980-1990* ∆GOVEXP -0.063 [0.512] -0.050 [0.640] R² 0.001 0.005 0.040 0.042 0.354 0.364 0.382 0.385 Adjusted R² -0.026 -0.050 -0.013 -0.041 0.209 0.166 0.097 0.027 Skewness (SE) -1.615 -1.569 -1.317 -1.293 -0.956 -0.914 -1.143 -1.103 Kurtosis (SE) 5.314 5.181 4.617 4.568 4.228 3.970 4.408 4.184 Jarque-Bera (SE) 25.642 [0.000]*** 23.737 [0.000]*** 15.516 [0.000]*** 14.867 [0.001]*** 8.385 [0.015]** 6.958 [0.031]** 11.711 [0.003]*** 10.184 [0.006]*** White (F-stat.) 0.586 [0.562] 0.579 [0.680] 1.659 [0.172] 1.113 [0.382] 1.360 [0.252] 1.381 [0.242] 1.385 [0.235] 1.102 [0.404] Arch (F-stat.) 0.686 [0.725] 0.695 [0.717] 0.438 [0.908] 0.438 [0.908] 0.459 [0.895] 0.625 [0.774] 0.381 [0.939] 0.483 [0.880] Durbin-Watson 1.579 1.570 1.676 1.664 2.078 2.137 2.052 2.086 Ramsey Reset quadratic (F-stat.) 1.306 [0.261] 0.200 [0.658] 0.010 [0.922] 0.083 [0.775] 0.247 [0.623] 8.441 [0.007]*** 0.026 [0.873] 0.619 [0.439] Notes:
◦ Please refer to Table 1. for variable definitions. The corresponding p-value is provided between brackets; *
significance at < 0.10 level; ** significance at < 0.05 level; *** significance at < 0.01 level.
◦◦The sample for OECD countries consists of 39 recessions.
◦◦◦ In order to avoid perfect multicollinearity, the intercept and slope dummy variables for Australia, Africa and the
