The financial crisis that started in the mortgage sector of the United States in 2007 turned into a worldwide credit crunch and subsequently triggered a global recession in 2009. With access to credit markets hampered by financial distress, private consumption falling owing to income and wealth effects, and new investment constrained by the negative economic prospects, governments implemented numerous measures to restore growth and regain market confidence (IMF, 2009a). Governments’ policy reactions have focused on fixing the banking system to help reestablish the flow of credit to the economy and implementing fiscal and monetary stimulus packages to sustain aggregate demand and prevent a downward spiral of output (IMF, 2009d). As room for monetary easing rapidly shrank, reflecting limited space for additional interest rate cuts and impaired monetary policy transmission channels, fiscal policy became the principal tool for stimulating economic recovery (Christiano, Eichenbaum, and Rebelo, 2009). To what extent fiscal policy will be effective in supporting growth recovery both in the short term and over time is subject of much debate (Jansen et al., 2008).
Countercyclical fiscal policies—comprising discretionary budget measures and the operation of automatic stabilizers—have generally helped shorten recession spells in advanced economies during previous crisis episodes (IMF, 2009b). The evidence is more mixed in emerging market economies where procyclical spending bias, narrow automatic stabilizers, and limited credit access have constrained governments’ ability to provide fiscal stimulus during adverse economic periods (Kaminsky, Reinhart, and Vegh, 2004). Initial fiscal conditions generally play a key role in crisis responses (Alesina et al., 2002) in both advanced and emerging economies. Countries are more likely to adopt countercyclical fiscal policies if sufficient fiscal space was created before the crisis. The success of fiscal policy in restoring growth also depends on the role of accompanying macroeconomic policies and on the design of the fiscal stimulus packages, as the size of multipliers varies across government spending and tax measures.
One of the key findings of the literature is that fiscal responses lead to sustained economic recoveries after the crisis only when financial sector’s vulnerabilities are addressed without endangering fiscal sustainability (IMF, 2009a). Crisis resolution measures generally entail costly government restructuring of private sector’s balance sheet, including of the financial sector, which can have a lasting negative impact on public debt levels. Furthermore, government interventions to boost private sector credit and domestic demand could leave the economy exposed to the risk of high-inflation and lower private investment growth.
Therefore, there is a potential conflict between the size of countercyclical fiscal expansions during downturns and their medium-term growth implications.
Against this backdrop the contribution of this paper is twofold. First, we focus on crisis episodes originating in the banking sector, which are of systemic nature (Laeven and Valencia, 2008), to assess the effectiveness of fiscal policy in restoring growth during distress times and sustaining economic expansion in the post-crisis period. While studies have been carried out to assess the role of policy responses during recessions (Claessens, Kose, and Terrones, 2008; IMF, 2009b), detailed evidence on the fiscal policy effects during financial distress periods is lacking. During financial crises, the environment for fiscal policy implementation is made more difficult by the high economic cost associated with the shock. Moreover, financial distress can lead to capital market freezes that make it difficult to access financing for deficit expansions.
Second, we focus on the composition of fiscal policy response to assess its effectiveness during shocks. The composition of government fiscal expansions and its impact on crisis length and post-crisis output recovery have not been dealt with in sufficient detail in the literature. However, one could expect fiscal policy composition to play a key role in determining both the likelihood of exiting a crisis and the medium-term growth prospects, as fiscal multipliers differ across fiscal policy instruments. Moreover, tax and spending measures adopted during financial distress periods can have long-term implications for economic efficiency and productivity growth when the crisis is over (Gali, Lopez-Salido, and Valles, 2005; Ghosh et al., 2009; Rogoff and Reinhart, 2009).
Therefore, the objective of this paper is to answer the following questions:
These questions have not been addressed in the literature, mainly because of lack of comparable fiscal data and difficulties in defining financial crisis episodes. To overcome this problem, we use a recently constructed database on financial crises (Laeven and Valencia, 2008) to assess the efficacy of fiscal policy during these episodes. This database comprises over 100 banking crisis episodes that occurred in the world between 1980 and 2008.
We find that fiscal expansions shorten the duration of these crises. The composition of the fiscal expansion package is, however, key to its success. Public consumption is more effective than public investment in reducing the duration of downturns because of its timely impact on aggregate demand, while cutting consumption taxes is correlated with shorter crises than income tax reductions as the impact of tax reductions on consumers is more widespread. While countercyclical fiscal expansions have no effect on post-crisis output growth, the composition of fiscal policy responses matters for economic recovery: increasing the share of public investment during shock periods is an effective way for improving medium-term output performance, while government consumption has no significant effect. Cutting the share of income taxes removes distortions that hamper medium-term economic growth, while consumption tax reductions during crises undermine future economic performance. These findings point to a potential trade-off in the use of fiscal policy instruments between short-term and medium-term growth objectives: a result not yet highlighted in the literature. They also stress the importance of fiscal response composition. Insufficient fiscal space and public debt sustainability concerns can, however, limit the effectiveness of fiscal expansions during crises.
The remainder of this paper is organized as follows: Section II reviews the relevant literature. Section III describes the data and the econometric approach. Section IV presents the empirical results followed by robustness tests in Section V. The concluding section summarizes the results and discusses the key policy implications.
II. LITERATURE REVIEW
Until recently, the study of financial crises has typically focused either on historical experiences of advanced countries (mainly the banking panics before World War II), or on more recent episodes in emerging market countries. An important strand of this literature deals with the controversial issue of identifying and classifying different types of episodes that occurred in the last century. There are two major references in this area.
First, Reinhart and Rogoff (2008a, 2008b, 2009) mark banking crises as two types of events: bank runs that lead to the closure, merger, or takeover by the public sector of one or more financial institutions; and if there are no runs, the closure, merger, takeover, or large-scale government assistance for an important financial institution that marks the start of a string of similar outcomes for other financial institutions. With these criteria, they identify 66 cases that occurred between 1945 and 2007. They find that banking crises lead to sharp declines in tax revenues, as well as to significant increases in government spending. On average, they find that government debt rises by 86 percent during the three years following a banking crisis, and at the end of this period, growth resumes slowly to reach an average annual rate of 2½ ercent in the third year after the crisis.
The second major reference is the paper by Laeven and Valencia (2008), which introduces a new dataset on banking crises, with information on the type of policy responses implemented to resolve these crises in different countries. Under their definition, in a systemic banking crisis, a country’s corporate and financial sectors experience a large number of defaults and financial institutions and corporations face difficulties repaying loans on time. Using this mix of objective data and subjective assessments,4 they identify 124 systemic banking crises over the period 1970–2007, and estimate that fiscal costs net of recoveries associated with these crises average about 13 percent of GDP, while output losses average 20 percent of GDP.
Many authors have also focused on the origins of banking crises. These studies have typically found that crises tend to erupt when the macroeconomic environment is weak, particularly when growth is low and inflation and interest rates are high (Demirgüc-Kunt and Detragiache, 1998; Collyns and Kincaid, 2003). Others have focused instead on the consequences of these crises, including the study by Reinhart and Rogoff (2009) cited above Claessens, Kose, and Terrones (2008) took the analysis one step further and studied recessions caused by credit contractions, those associated with house price declines, and episodes of equity price declines. Their results show that the interaction between macroeconomic and financial variables can play major roles in determining severity and duration of recessions. Specifically, they find evidence that recessions associated with credit crunches and house price busts tend to be deeper and longer than other recessions.
The analysis of policy responses to these crises constitutes another area of interest for scholars. Some studies have analyzed the type of containment and resolution policies aimed at stabilizing the banking sector during financial crises (Laeven and Valencia, 2008). Others have assessed the macroeconomic policy response. Claessens, Kose, and Terrones (2008) and IMF (2009b) find that both monetary and fiscal policy tend to be countercyclical during recessions, credit contractions, and asset price declines. In these episodes, fiscal policy appears to be more accommodative, suggesting a more aggressive countercyclical fiscal stance. They also find that expansionary fiscal policy (proxied by discretionary government consumption) tends to shorten the duration of recessions. The lessons from these analyses have stimulated other papers with a more prescriptive approach. For instance, one paper argues that an optimal fiscal package to mitigate the adverse consequences of financial crises should be large, lasting, diversified, contingent, collective, and sustainable (Spilimbergo et. al, 2008).
Finally, the increase in fiscal deficits and public debt linked to fiscal policy expansions during crises have also led to a discussion of the perception of financial markets about fiscal sustainability. Ardagna (2009) shows that financial markets value fiscal discipline, since interest rates on long-term government bonds and stock market prices worsen considerably in periods of fiscal expansion. Looking at the composition of fiscal policy, Akitoby and Stratmann (2008) show that financial markets react to the composition of the budget in emerging market economies. For example, revenue-based adjustments lower government spreads more than expenditure-based ones, and debt-financed spending increases sovereign risks. Baldacci, Gupta, and Mati (2008) find that the composition of fiscal policy matters for government spreads, but debt levels matter as well. They show that spending on public investment contributes to lower government bond spreads, as a long as the fiscal position remains sustainable and the fiscal deficit does not worsen.
Our paper builds on the above literature to assess the relationship between the composition of fiscal policy response during banking crises, duration of these episodes, and post-crisis economic performance. While Laeven and Valencia (2008) report multiple measures of containment and resolution policies, they only use one measure of fiscal policy (the budget balance) and their work is purely descriptive, without causal analysis. Subsequent empirical work (IMF, 2009b; 2009c) also proxies the fiscal policy response using government consumption and primary balance indicators. Instead, we measure the effectiveness of fiscal policy in terms of the different budget categories (both on the revenue and spending side) and the observed characteristics of each episode.
III. FISCAL POLICY DURING BANKING CRISES
This section describes the impact of banking crises on budgets. We build a dataset of banking crises from a panel of 182 countries between 1980 and 2008. We follow the criteria established by Laeven and Valencia (2008) and identify 118 episodes of banking crises that occurred in 99 different countries (in some countries up to four times during the period, such as in Argentina). We complement Laeven and Valencia´s database with additional data from the World Economic Outlook (WEO), the Government Financial Statistics (GFS), and the Global Financial Database (GFD).
Unlike Laeven and Valencia (2008), we not only identify the start of the crises, but also define their duration. We are aware of the difficulties in identifying the duration of banking crises, since there is no single financial indicator that is valid for all of them. Nevertheless, regardless of the origins and the characteristics of each banking crisis, we assume that a crisis ends after two consecutive years of real GDP growth above ½ percentage points per year. For the purpose of this paper this definition allows us to link the crisis duration with the negative output implication of the crisis. This is consistent with the focus on the effects of fiscal policy responses in restoring economic stability. In Section IV, we test the robustness of our results to a different definition of crisis duration, based on stock market performance.
Using the above criteria, we find that banking crises lasted on average for 2½ years, with 85 percent of the crisis episodes lasting between one to four years, and only one episode lasting eight years (see Figure 1). This is consistent with the findings of Claessens, Kose, and Terrones (2008) who report an average duration of recessions linked to credit crises of 2½ years. Reinhart and Rogoff (2008b) estimate an average duration for their reduced sample of financial crises of about three years.
Consistent with previous studies, we also find that banking crises generate large economic costs. Peak-to-trough figures show that the average GDP growth rate fell by more than 5 percentage points during the crisis, general government debt increased by 39 percentage points of GDP and the budget deficits increased by 6.9 percentage points of GDP.
To assess the behavior of fiscal variables during crises episodes and in their aftermath, we follow the recent literature (Ardagna, 2009) and calculate the overall change in the variables: (i) in two years prior to the start of the crisis; (ii) during the crisis; and (ii) in the two years after the crisis. Results of descriptive statistics are expressed as a percentage of GDP and as a percentage of total revenues or total expenditures.
During banking crises, fiscal deficits increased by almost 6 percentage points (more than 2 percent of GDP per year) and public debt worsened by 27 percentage points of GDP (about ⅓ of the preexisting average debt level, which was on average 78 percent of GDP). Total revenues fell considerably during the crisis period (more than 3½ percentage points of GDP) and government expenditures rose by more than 2 percentage points of GDP.
Social contributions also fell considerably in the period, accounting for about ⅓ of the total decline in public revenues. After the crisis, revenue collection improved, in particular, taxes related to the economic recovery and the associated improvement in private income and profits. In terms of expenditure, there was a significant increase in current expenditure during banking crises. Interest payments, transfers, and government’s purchase of goods and services show the sharpest increase. The rise in public sector salaries is weaker and other expenses fall slightly as a percentage of GDP. Public investment remains broadly unchanged during the crisis, but recovers significantly after the crisis, more than offsetting the decline in other spending items.
IV. THE EFFECTIVENESS OF FISCAL RESPONSE
This section assesses the effectiveness of fiscal policy response in: (i) reducing the duration of banking crises and (ii) promoting economic growth following a crisis. The previous section showed that during banking crises fiscal deficits widened, mainly because of an increase in public consumption, a freeze in public investment, and a fall in revenue from income taxation and international trade. This outcome reflects the operation of automatic stabilizers and incorporates the effects of discretionary fiscal policy changes implemented by governments in response to output declines (Gali, 1994).
In a standard Keynesian framework, we would expect a fiscal expansion driven by cuts in taxes and increases in public spending to shorten the duration of the crisis and sustain medium-term growth. Higher government spending and lower taxes help boost aggregate demand during downturns associated with banking crises, replacing falling private consumption as a growth engine (Arreaza, Sorensen, and Joshua, 1999). Public investment measures can, at least in part, offset the collapse in private investment (Aschauer, 1989). A simple plot of changes in levels of these variables as a ratio to GDP against the duration of banking crisis episodes supports these hypotheses. Shows a strong positive correlation between higher deficits and shorter crisis duration. However, budget composition changes matter as well as the size of the fiscal package. Higher public consumption (as a percentage of total expenditures) and lower income taxes (as a percentage of total revenues) also shorten the duration of banking crises. The contribution of public investment in reducing the crisis length is, however, significantly weaker. This result is somewhat surprising in light of the relative size of estimated fiscal multipliers for various tax and spending measures which point to larger multipliers for public investment than government consumption (Spilimbergo, Symansky, and Schindler, 2009). However, issues related to the timeliness of disbursements matter: while government consumption has an immediate impact on aggregate demand through the direct purchase of goods and services by the government, public investment may affect the economy with a delay, as procedural bottlenecks and lack of shovel-ready projects may slow down project execution.
An increase in the share of public investment during the crisis significantly raises post-crisis GDP growth and this increase is more than that brought about by a higher share public consumption in the budget. The most likely reason behind this result is that public investment, particularly in infrastructure, can raise productivity while government’s current consumption may crowd out private consumption over time. Reducing income taxation during crises is also beneficial for output growth following the crisis, as the distortionary impact of high tax burdens is mitigated. This does not hold for taxes on goods and services; their positive impact on private consumption is more than neutralized by negative expectations of future higher taxes to finance growing fiscal deficits. In the next step, we use a multiple regression framework to test if the above relationships hold when other covariates of crisis length and output growth are included in the model specification. Along with the budget balance (in percent of GDP), we also use a dummyvariable indicator of large fiscal expansions during the crisis episode to capture only major changes in fiscal policy.
To build this indicator we follow Laeven and Valencia (2008) and create a variable labeled “expansionary fiscal policy” that takes value equal to 1 if the budget balance worsens by more than 1½ percent of GDP in the first three years following the onset of the crisis, and is equal to zero otherwise.
V. ROBUSTNESS ANALYSIS
This section assesses the strength of the above results on the basis of three robustness analyses:
This paper assessed the effects of fiscal policy responses during 118 episodes of systemic banking crises in advanced and emerging market economies. The results indicate that timely countercyclical fiscal responses (both due to discretionary measures and automatic stabilizers), accompanied by actions to deal with financial sector weaknesses, contribute to shortening the length of crisis episodes. During crisis caused by financial sector distress, fiscal expansions increase the likelihood of earlier exit from a shock episode. Expansionary fiscal policies reduced the crisis duration by almost one year. These results hold for different definitions of crisis duration and alternative specification and estimation methods. The findings are consistent with recent studies that highlight the importance of countercyclical policy in response to recessions associated with financial sector problems (Classens, Kose, and Terrones, 2008; IMF, 2009b; IMF, 2009c).
Initial fiscal conditions matter for fiscal performance during shocks. In countries with high precrisis ratios of public sector debt to GDP, lack of fiscal space not only constraints the government’s ability to implement countercyclical policies, but also undermines the effectiveness of fiscal stimulus and the quality of fiscal performance. In countries with high debt, crises lasted almost one year longer. The effect of high public debt on duration completely offset the benefits of expansionary fiscal policies in these countries. Similar results are found for countries with lower per capita income, as poor implementation capacity and high macroeconomic risks limit the scope and the effects of fiscal expansions during crises (Botman and Kumar, 2006). These findings point to the importance of creating fiscal space and enhancing macroeconomic stability in tranquil times to limit the risk of falling into crises and to enhance the effectiveness of policy responses when exogenous shocks hit countries (Tavares and Valkanov, 2001). In emerging market economies, attention needs to be paid to strengthening fiscal institutions, reduce political risks and improve budget execution capacity to reap the benefits of countercyclical fiscal policies (Baldacci, Gupta, and Mati, 2008).
The composition of fiscal expansions matters for crisis length—a point that has not been studied in the literature. Stimulus packages that rely mostly on measures to support government consumption are more effective in shortening the crisis duration than those based on public investment. A 10 percentage point increase in the share of public consumption in the budget reduces the crisis length by three to four months. Reducing the share of income taxes is less effective than consumption taxes in shortening the length of a banking crisis. These results suggest that tailoring the composition of fiscal response packages is important for enhancing the effectiveness of countercyclical fiscal measures in both advanced and emerging market economies (Spilimbergo et al., 2008; IMF, 2009).
Fiscal expansions do not have a significant impact on output recovery after the crisis though. Crises can have long-term negative effects, damaging human and physical capital with negative implications for productivity and potential output growth. Early recovery from a crisis is therefore important, to minimize output losses in the short term and enhance medium-term growth prospects. This calls for timely fiscal responses during downturns. However, fiscal policy responses may not be effective when initial fiscal conditions are poor and fiscal space is limited. High public debt levels and past macroeconomic instability limit the scope for countercyclical deficit expansions and hamper the effectiveness of fiscal stimulus measures as markets perceive the higher future fiscal risks entailed by larger deficits (Balduzzi, Corsetti, and Foresi, 1997; Uribe, 2006).
The quality of the fiscal stimulus package matters most for post-crisis growth resumption, with fiscal responses relying largely on scaling up the share of public investment in the budget showing the largest positive effect on medium-term output growth. A one percent increase in the share of capital outlays in the budget raised post-crisis growth by about ⅓ of one percent per year. Income tax reductions are also associated with positive growth effects.
The results of the short-term and medium-term impacts of fiscal policy during financial crises highlight a potential trade-off between short-run aggregate demand support measures and medium-term productivity growth objectives in fiscal policy response to shocks. Implementation lags for government investment, which were documented also during the current crisis, may be, at least in part, responsible for these results. They also point to careful consideration of the composition of fiscal stimulus packages, as different short-term and medium-term fiscal multipliers can affect fiscal policy performance during the crisis and in its aftermath (Spilimbergo, Symansky, and Schindler, 2009).
The results of the paper also call for further research. Economic theory predicts that, in normal circumstances, fiscal expansions tend to crowd out private investment and increase the cost of financing for the private sector. However, the empirical findings presented here indicate that an increase in the share of public investment (as a percentage of total public spending) is compatible with an increase in the share of private investment (as a percentage of total investment) during banking crises, and both can have a positive contribution to long-term growth in the subsequent period. This constitutes a very preliminary evidence of the crowding-in effects potentially attributed to fiscal policy in situations of financial stress (Aschauer, 1989). But a proper test of this hypothesis was beyond the scope of this paper.
The recent debate on the link between austerity and growth has focused on the short run. This column discusses recent research into the link between fiscal consolidation and medium-term growth under different financial conditions. If credit is not available to consumers and investors, private demand is less able to compensate for cutbacks in public demand, so large spending cuts can have a negative effect on growth. Difficult financial conditions probably explain why fiscal adjustments that worked in the 1990s have not produced similar beneficial effects on growth in recent years.
In the aftermath of the recent financial crisis, the discussion of the effects of fiscal adjustment on economic growth has intensified. While some scholars have focused on the characteristics of the fiscal consolidation needed to bring public debt down from historically high levels, others have examined the effects of alternative strategies on economic performance. The VoxEU debate aptly covered in “Has Austerity Gone Too Far?” (Corsetti 2012) sums up the conflicting positions. Most scholars have delved into different dimensions of fiscal adjustments, including their timing, composition, and duration. The empirical analysis of the growth effects of these policies, however, has largely concentrated on the short run. This has left unanswered the question of how fiscal consolidations affect growth over the longer term.
In our view, important insights are gained from taking a medium-term perspective of fiscal adjustment and its effects on growth. In particular, the composition of fiscal adjustment, and its interaction with initial and accompanying conditions, has important consequences for economic activity. Monetary policy, exchange rates, and financial conditions are important factors shaping the medium-term effects of fiscal consolidations on economic performance (see Cottarelli 2012 and Buti 2014).
In this context, we would like to draw attention to another dimension – the interplay of credit constraints, the composition of fiscal adjustment, and growth. This is particularly relevant in the current environment, where the legacy of the financial crisis has constrained credit supply and affected the channels through which fiscal policy impacts growth.
Debt-reduction episodes under different financial conditions
In a recent paper (Baldacci 2013), we show that there is a link between fiscal adjustment and medium-term output growth in the presence of credit constraints. We use a database of 79 debt-reduction episodes achieved by means of fiscal adjustment, covering advanced, emerging, and developing economies between 1980 and 2012. In particular, we focus on the relationship between the quality of adjustment (measured by the relative contribution of cuts in current spending to total deficit reduction) and average economic growth in the medium term. In this context, we define the medium term as the five years after the debt-reduction episode has ended. The relationship between the quality of adjustment and medium-term growth is weakly positive under normal conditions in our sample. However, this relationship becomes negative when credit constraints and bank deleveraging conditions are taken into account.
A shortage of credit and impaired financial channels can damage growth, while spillovers of risks from the financial sector to sovereign debt markets can affect public debt sustainability. While these figures are only illustrative, in the paper we perform a robust empirical analysis in which we interact traditional variables that measure the size and the composition of fiscal adjustments with indicators that measure the relative health of the financial sector. In particular, we focus on two financial variables: the growth of credit to the private sector and bank deleveraging, with the latter measured by the change in the capital-to-assets ratio.
We find that if credit is not available to consumers and investors, private demand has more difficulties in compensating for cutbacks in public demand. As a consequence, strong spending cuts can have a negative effect on medium-term output growth; this impact is larger than in traditional models that do not take credit constraints into account. When countries start the adjustment with high unemployment, the negative effect of spending cuts on growth is even more acute.
The combination of low credit growth, bank deleveraging, and public debt consolidation can change the way economic agents assess the effects of government policies. In particular, the fiscal mix (that the mix of expenditure and revenue policies) that under normal circumstances would have delivered growth-boosting public debt reductions may not be successful in an environment of credit restrictions and uncertainty about the future. The presence of troubled financial conditions probably explains why fiscal adjustments that worked in the 1990s have not produced similar beneficial effects on growth in recent years.
These results are consistent with those of Eggertsson and Krugman (2010), who illustrate the adverse growth consequences of deleveraging when the effectiveness of monetary and fiscal policy is constrained by the liquidity trap. They also confirm that the deleveraging of both the public and the private sector at the same time has a negative impact on growth (Bornhorst and Ruiz Arranz 2013).
The policy implications of these results are significant – when bank deleveraging is high and credit is not flowing to the private sector, public debt consolidations should be gradual. They should also be based on an appropriate combination of revenue and expenditure measures rather than spending cuts alone (IMF 2013).
In a nutshell, credit conditions matter for the medium-term success of fiscal adjustments, as judged by their impact on growth. The expenditure-based consolidations of the past, which worked well in an era without credit constraints, may disappoint when financial conditions are less favourable. Fiscal adjustments based on a prudent mix of both revenue and expenditure measures may be a better recipe for success at the current juncture.
Bornhorst, Fabian and Marta Ruiz Arranz (2013), “Indebtedness and Deleveraging in the Euro Area”, 2013 Article IV Consultation on Euro Area Polices: Selected Issues Paper, IMF Country Report 13/232, Chapter 3.
Eggertsson, Gauti B and Paul Krugman (2010), “Debt, Deleveraging and the Liquidity Trap: A Fisher-Minsky-Koo Approach”, mimeo.
IMF (2013), “Reassessing the Role and Modalities of Fiscal Policy in Advanced Economies”, IMF Policy Paper.
A better method is to look at changes in the cyclically adjusted primary budget balance—ie, the surplus or deficit after stripping out interest payments and temporary effects of the economic cycle. Isolating temporary effects is not an exact science, but the OECD, a club of mostly rich countries, has had a go. The change in this measure, from the point when public spending was at its most profligate to the moment when it was most restrained (or the projected balance for this year if belt-tightening continues), provides a fairer measure of austerity (see chart).
Portugal, Ireland, Italy, Greece and Spain—the PIIGS, as investment bankers’ shorthand has it—were in the direst fiscal straits in the crisis and, naturally, have been the most austere since. Italy has reduced its underlying primary deficit by 4.7% of GDP; the others, by more than 8% of GDP. These figures are huge: 8% of GDP is equivalent to average government spending on pensions in the OECD. No one should accuse the Greek government, in particular, of not cutting back enough: the figures reveal tightening of a whopping 17.2% of underlying GDP between 2009 and 2015. At the other end of the scale, Germany has barely had to cut back at all, and in fact the OECD expects it to loosen its purse-strings slightly this year. No wonder the PIIGS have squealed.
Even this measure of austerity is not perfect, however. By measuring from the high point of profligacy, it includes one-off borrowing intended to inject life into slumping economies. For example, the apparent 6.4% improvement in America’s underlying primary balance rests in part on the expiry of a fiscal stimulus estimated by the IMF to be worth around 2% of GDP in 2009. Although withdrawing stimulus is painful, most would agree that this fiscal splurge in the base year makes governments appear to be donning a hairier shirt than they really are.
Cutting to stand still
The other caveat is that the measure obscures the distinction between countries that saw GDP growth and those that saw massive declines. When an economy is shrinking fast, even keeping spending flat as a share of GDP involves deep cuts in cash terms. Thus Greece has had to slash actual spending by more than a quarter to achieve an 11.2 percentage-point cut in spending as a share of GDP. The British government, in contrast, will have managed to reduce underlying spending, excluding debt interest, as a share of GDP by 3.2 percentage points, but economic growth has allowed it to achieve this by holding this measure of spending roughly constant in real terms (ie, after accounting for inflation).
Aggregate numbers mask other differences, too. Public-sector workers take little comfort from the knowledge that overall spending is buoyant if their salaries have been frozen while spending on social welfare has grown. The OECD’s estimates suggest that this is indeed what has happened: in America, Britain and the PIIGS, spending on public services has been cut relative to spending on benefits and pensions. In Portugal general government consumption (a broad measure of spending on public services) has been slashed by almost a fifth in real terms since 2009, whereas social-security spending has crept up by 4%. And even rising spending on social welfare may feel austere if ageing populations are putting pressure on pension systems.
From any perspective, however, the recent bout of belt-tightening looks severe. A published last year by Julio Escolano, Laura Jaramillo, Carlos Mulas-Granados and Gilbert Terrier of the IMF puts the cuts in historical context. The authors compiled a database of 48 austerity drives in rich countries between 1945 and 2012, all aimed at steadying public debt as a share of GDP. They find that around half of these consolidations amounted to 5% or more of GDP, and a quarter to 7.5% or more. Italy’s recent experience is about average, therefore, and Britain’s (so far) below par. But Greece, Ireland, Portugal and Spain have been far more austere than the norm.
Greece’s recent privations are the most severe of all those that the authors evaluated. Second place is also taken by Greece, which underwent a previous bout of austerity in 1990 to secure (you guessed it) membership of the euro. Germany’s fiscal retrenchment in 1996 earns fifth place. But that dose of Swabian spending restraint, which induced huge strikes, ultimately amounted to just 10% or so of GDP, a little over half of what Greece has endured since 2009.
Austerity has not been adopted at random. Those governments that have cut back the most were also those that spent most recklessly before. Greece may have tightened by 17% of GDP, but at its peak its underlying primary deficit was a clearly unsustainable 12%. Citizens of less spendthrift countries such as Germany are entitled to condemn the PIIGS’ past excesses. They may legitimately rail about the pace of structural reform. But they cannot denounce them for doing too little on the public finances.
Many countries around the world have accumulated large public debt in the aftermath of the global financial crisis.
Debt accumulation reflects revenue losses resulting from economic effects of the crisis, and to a lesser extent, exceptional fiscal stimulus and financial sector support measures (IMF, 2010). This is particularly the case in advanced economies, where debt is projected to rise from an average of about 73 percent of GDP at end-2007 to about 108 percent of GDP at end-2015. This level of debt was only reached after World War II (Cottarelli and Schaechter, 2010). Moreover, the debt surge is occurring at a time when pressure from age-related spending (e.g., pensions and health care) is building up in many countries (IMF, 2009d). Public debt has also increased in some emerging economies (e.g., in Central and Eastern Europe) during the recession, although the bulk of these economies have not been hit as hard as advanced economies, reflecting their relatively healthier fiscal positions before the crisis. Nonetheless, emerging economies tend to have a lower debt tolerance, owing to narrower and more volatile revenue bases, and are exposed to spillover from solvency risks in advanced sovereigns.
The increase in fiscal vulnerabilities following a banking crisis has been highlighted by the literature. Studies have shown that banking crises have large fiscal consequences (Freydl, 1999; IMF, 2009a; 2009b; 2009d; Laeven and Valencia (2008); and Rogoff and Reinhart (2009)). For example, Rogoff and Reinhart found that government debt on average rose by 86 percent in the three years following a banking crisis in a sample of historical episodes, while Laeven and Valencia estimated that the average fiscal cost of banking crises (net of recoveries) was slightly less than 15 percent of GDP in the last three decades. More recently, Baldacci, Gupta, and Mulas-Granados (2009) reported that in the period covered by Laeven and Valencia the average peak-to-trough increase in the public debt-to-GDP ratio was about 40 percentage points.
Higher public debt raises solvency risks, constrains the capacity to use fiscal policy as a countercyclical tool, and can increase borrowing costs for sovereigns (IMF, 2009d). Ultimately, the increase in public debt may reduce output growth and productivity (IMF, 2010). As economies recover from the recession, therefore, the challenge for governments is to regain fiscal health through budget consolidation and pro-growth structural reforms to reduce public debt.
However, reducing high levels of public debt after the recent crisis can be challenging. The legacy of the crisis on potential output and the unprecedented simultaneous increase in public debt levels worldwide make this effort particularly demanding (Alesina and Ardagna, 2009). Lowering high public debt to a sustainable level calls for large improvements in the structural primary balance and a favorable dynamic of the growth- interest rate differential. The latter could be difficult to achieve in the aftermath of a global banking crisis. This is consistent with the evidence on previous instances of postcrisis debt reduction. In the Baldacci, Gupta, and Mulas-Granados study, only 12 percent of the countries were able to reduce their debt to precrisis levels 16 years after the end of the crisis.
Against this background, the questions that motivated this study are:
We apply survival analysis to a sample of 100 banking crisis episodes that occurred between 1980 and 2008. At the outset, we study patterns of postcrisis debt stabilization in the sample. We then carry out a parametric analysis to assess the determinants of successful debt reductions after a banking crisis. In particular, we investigate the role of fiscal policy composition, taking into account accompanying monetary policy and changes in other economic variables during the consolidation period.
Our baseline results define a prudent debt level as an “absolute debt target.” As returning to the precrisis public debt level may prove insufficiently ambitious for countries that had high debt ratios prior to the crisis, we define specific thresholds of 60 percent of GDP for advanced economies and 40 percent of GDP for emerging market economies reflecting the perceived higher risk for them. This is consistent with the literature on fiscal sustainability, which derives targets for the ratio of public debt to GDP to measure success, independently of the debt level before the crisis (Cottarelli and Viñals, 2009; IMF, 2010). The 60 percent of GDP target for advanced economies is also the median debt-to-GDP ratio in advanced economies in the precrisis period.
We use three definitions of success:
We complement this analysis with various robustness tests.We examine the impact of alternative debt targets and definitions of success to ensure that our results do not depend on these assumptions. The remainder of this paper is organized as follows: Section 2 reviews the relevant literature. Section 3 describes the dataset used in the analysis. Section 4 reports the findings of the parametric analysis, and Section 5 discusses the robustness of these results. The concluding section summarizes the findings and discusses their policy implications.
II. LITERATURE REVIEW
The success of fiscal adjustment strategies in reducing public debt is influenced by the size of fiscal consolidation. Giavazzi and Pagano (1996), and Giavazzi, Jappelli, and Pagano (2000) find that large consolidations are important for success. They show that sizable improvements in the structural primary balance signal a regime change, and thus have a positive impact on private sector expectations and consumption behavior. When initial debt is high, as in most advanced and some emerging market economies today, a credible and sustained improvement in the primary balance can spur economic growth, which in turn can help achieve faster debt reduction (IMF, 2009d).The growth spurt stems from a reduction in distortions associated with high levels of taxation to finance elevated debt levels, the wealth effect enjoyed by consumers expecting a decline in the future tax burden, the increase in labor supply from lower taxes and spending (Alesina and Ardagna, 2009), and the beneficial effect of lower interest rates on capital accumulation triggered by public debt reduction, particularly when global sovereign bond supply is expanding (Baldacci and Kumar, 2010).
The composition of the adjustment is also relevant. Alesina and Perotti (1995, 1996); McDermott and Wescott (1996); Alesina, Perotti, and Tavares (1998); Alesina, Ardagna, Perotti, and Schiantarelli (1999); and Alesina and Ardagna (2009) show that expenditure cuts increase the likelihood of reducing public debt, in particular when these cuts are concentrated on public transfers (e.g., pensions, subsidies and other entitlements) and government wages. They also find that the composition of fiscal adjustment is more important than its size in reducing the stock of public debt and generating expansionary effects on output. In these studies, spending cuts are found to be more likely than tax increases to stimulate output growth during the fiscal adjustment period.
Sustained fiscal consolidation efforts tend to be associated with successful debt reductions. This result has been stressed in many studies, including by Von Hagen, Hallett, and Straucht (2001), Maroto and Mulas-Granados (2002, 2007), and Gupta, et al. (2005). More persistent adjustment efforts signal the authorities’ commitment to debt consolidation Many authors (e.g., Perotti (1999); Von Hagen, Hallett, and Straucht (2001); and Lambertini and Tavares (2001)) have emphasized the role that positive initial fiscal conditions, initial economic growth, monetary conditions and exchange rate devaluations play in the likelihood of success during fiscal adjustment episodes.
Decisions regarding the timing, the duration, the size, and the composition of adjustments are usually subject to institutional and political constraints. Among them, the influence of the cabinet’s ideology on fiscal policy (Perotti and Kontopoulus, 2002; Mulas-Granados, 2003, 2006); the impact of the electoral system and the budget process (Persson and Tabellini, 1999; Hallerberg and Von Hagen, 1997; Von Hagen, Hallett, and Straucht, 2001), and the proximity of elections (Alesina, Cohen, and Roubini, 1992; and Mulas-Granados, 2002 and 2007; Buti and Van den Noord, 2003) are important. More stable governments and stronger fiscal institutions (including well-designed fiscal rule frameworks) are more likely to remove the deficit bias in fiscal policy and can help achieve lower debt (IMF, 2009e). In contrast, high fragmentation in decision-making can have negative implications on the budget, making fiscal adjustment more difficult and eventually leading to higher debt (Roubini and Sachs, 1989; Grilli, Masciandaro, and Tabellini, 1991; and MulasGranados, 2003).
However, the above literature has not focused on the challenge of regaining debt sustainability after banking crises. On the one hand, it would be expected that most factors identified in the literature on effective fiscal consolidations are at work in the aftermath of banking crises. On the other hand, unwinding of large debt accumulated after these crises could require a different fiscal policy mix in light of the uncertainty about growth and financial sector health that typically follows these episodes and the need for large adjustment. This is compounded in the case of the current crisis by the twin challenge of fiscal adjustment and increased debt rollover risks across a number of countries in the world, which may have negative implications for interest rates and make debt reduction more difficult to achieve. Furthermore, it is now recognized that policy response to financial crises is critical in reducing crisis length and increasing postcrisis growth (Baldacci, Gupta, and Mulas-Granados, 2009) and this could affect the likelihood of public debt reduction.
The following sections explore the probability of achieving successful fiscal consolidations as defined above by using survival analysis.
III. DESCRIPTIVE ANALYSIS
In this paper, we use the sample of banking crises compiled by Laeven and Valencia (2008). They define banking crises as periods in which a country’s corporate and financial sectors experience a large number of defaults and financial institutions and corporations face great difficulties in meeting contractual obligations. Using a mixture of objective data and subjective assessments, they identify 124 systemic banking crises over the period 1970– 2008. We dropped 24 of them due to insufficient fiscal data in the WEO and GFS databases. As a consequence, our sample covers 100 banking crises in 99 advanced and developing countries.
There are two alternative methods to analyze the evolution of public debt in the aftermath of banking crises. The first approach entails looking at the change in debt-toGDP ratios after the crisis and describing debt reduction characteristics (e.g., timing, size, and composition). A second option is to define a threshold to identify successful debt consolidations cases and describe the features associated with these episodes. We use both approaches in the following sections.
A. Postcrisis Debt Trends
First, we analyze changes in the debt-to-GDP ratio in our dataset. In 65 out of 100 episodes, public debt levels at the end of the banking crisis were higher than before the crisis started. This number rises to 89 episodes if the debt level three years after the start of the crisis is compared to debt before the crisis. Overall, almost 90 percent of episodes left a legacy of higher debt.
About a decade after the end of the crisis, debt ratios fell only in 59 percent of the cases. However, in 40 percent of these episodes, debt was still higher than in the immediate aftermath of the crisis. Postcrisis fiscal consolidation lowered debt by up to 20 percentage points of GDP in 56 percent of the cases, but in more than a quarter of episodes, governments managed to reduce public debt between 20 and 40 percentage points of GDP. Larger consolidations were associated with high initial levels of debt as adjustment needs became more pressing in these countries. Larger debt reduction is also correlated with more long-lasting fiscal consolidation efforts, expenditure-based adjustments, and the contribution of tax revenues to total public revenues.
B. Survival Analysis Next
we turn to episodes of successful and unsuccessful debt consolidation. We generate a dummy variable that takes value zero every year after an episode of banking crisis in which public debt level (as a percentage of GDP) is above the “absolute target” (60 percent of GDP for advanced economies and 40 percent of GDP for emerging markets). This variable takes value of one when public debt is equal or lower than the threshold. Using the years in which debt remains above the threshold, we create a new variable called “Duration” that represents the length of the (successful) debt consolidation period. We generate 972 observations (net of missing data points), distributed among 100 episodes of banking crises. The minimum length of a debt-recovery episode is one year and the maximum is 24 years.
It is worth noting that in 57 percent of the episodes, debt was within the “absolute debt target” before the banking crisis started. In 62 percent of these episodes, debt rose beyond the target as a consequence of the crisis. This reflects both the impact on revenue of potential output losses as well the fiscal cost of the crisis-response measures.
Under the “absolute debt target” criterion, the average length of a successful debt recovery episode (that is, time taken to reach the target debt level) is about 10 years. This translates into a low probability of success for debt consolidations in the sample. As we relax the definition of success, the duration shortens. For example, the average duration of partially successful episodes (that is, a fall in debt equivalent to 50 percent of the required debt reduction to reach the target debt level) is slightly more than seven years and the average duration of limited success episodes (reducing debt by 10 percent of the required adjustment) is less than six years.
Under the “precrisis debt target,” the average length of a successful debt recovery episode is 9½ years. This is slightly less than the time needed to reduce debt to the 60/40 percent of GDP target. In episodes of partial/limited success, duration is seven years and six years, respectively. Reaching the absolute debt target “takes only half a year longer than reaching the” precrisis debt target. This implies that in the sample, precrisis debt levels were not much different from the prudent debt level.
The history of banking crises is full of contagion effects (which typically have a regional component) and reputation asymmetries (normally linked to the degree of economic development). This may be reflected in the duration of debt recovery episodes in the aftermath of banking crises. More importantly, public debt dynamics may follow different paths depending on the initial fiscal conditions, the size of the fiscal deterioration, and the budget composition during and after the banking crisis.
IV. PARAMETRIC SURVIVAL ANALYSIS
In this section, we assess the determinants of successful debt reduction in the aftermath of banking crises. We use a Cox regression model that is equivalent to the standard linear regression in the context of survival analysis; here the duration is measured in terms of the length of a successful debt reduction period. The probability of failing to reduce public debt within the target is regressed on a vector of determinants. This includes two sets of variables.
Results confirm the importance of the fiscal policy mix for successful debt reduction. ” These results can be summarized as follows:
There are two channels through which the fiscal policy mix affects the probability of successful fiscal adjustment. First, fiscal adjustments based on an appropriate combination of expenditure cuts and revenue increases allow countries to sustain persistent fiscal consolidations and larger debt reductions. This reflects the scope for large fiscal savings from the adoption of fiscal measures that improve the composition of the budget. Second, an appropriately balanced fiscal policy mix can boost output growth and help lower credit risk premia, thereby reducing the interest rate-growth differential component of debt dynamics.
Many countries around the world have accumulated large public debts in the aftermath of the recent banking crisis. As the economies recover from the recession, the challenge for governments is to regain fiscal stability by unwinding the exceptional fiscal stimulus when economic conditions permit and reducing public debt. The unprecedented simultaneous increase in public debt levels worldwide, however, makes this effort particularly demanding. This paper focused on factors that explain successful public debt consolidations following 100 episodes of banking crises during 1980–2008 using survival analysis. We find that debt consolidation is less successful when countries are hit by longer-lasting (and thus more severe) banking crises. This reflects higher uncertainty and permanent output losses resulting from these crises that make fiscal consolidation more challenging.
Successful debt consolidations are in general more likely when they are based on cuts in current expenditures. Accompanying policies are important; when monetary conditions are allowed to remain accommodative and risk premia are contained, debt reduction is more likely to be achieved: a key lesson for countries exiting the crisis and preparing to unwind fiscal and monetary support. This result also highlights the importance of credible fiscal adjustment strategies that anchor market expectations about fiscal sustainability. Lack of credibility can make debt reduction much harder to achieve and lead to potential selffulfilling expectations about rising solvency risks.
In contrast with the previous literature on fiscal consolidations, we find that raising tax revenues is important for debt reduction in countries with large consolidation needs. This reflects the large size of fiscal adjustment required in postcrisis periods and the need to maintain a balance between expenditure savings and revenue-raising measures to sustain the consolidation efforts long enough to bring debt under control. An appropriate fiscal policy mix would also help reduce economic inefficiency that hamper growth and would boost the credibility of fiscal consolidation thereby contributing to tighter credit risk spreads. However, higher taxation should not harm efficiency and has to minimize distortions, particularly in countries with high tax ratios. Simplifying the tax system by reducing excessive tax rates and broadening the tax base could help enhance revenue collection while shifting the burden of taxes from income and capital to consumption, pollution and property taxes could help reduce distortions (IMF, 2010).
Overall, fiscal adjustment can be complex in the aftermath of banking crises, requiring supporting actions to revive growth. In such circumstances, debt consolidations should rely on a combination of improvements in the primary balance and sustained economic growth. The former should be achieved by a combined strategy of revenue increases and expenditure savings while preserving productive investment. The latter requires implementation of structural reforms to enhance productivity as well as measures to reduce economic distortions in the economy. Improving the budget composition could be an additional important ingredient in the strategy to support growth by removing efficiency harming distortions and raising labor supply and savings.
Carmela Martín, Carlos Mulas-Granados* and Ismael Sanz
ABSTRACT: This article explores the spatial distribution of regional technology indicators in the EU over the last decade and its impact on cohesion. Thus, we find that public R&D spending and patent applications have converged among regions during the nineties. On the other hand, private R&D activities have diverged, as a result of an asymmetric expansion during the second half of the nineties. We show that when the dispersion of public R&D across regions diminished in the second half of the nineties, income disparities at regional level also decreased. Therefore, while technology policy based on efficiency criteria should remain as a policy tool for economic growth, this policy should be counterbalanced by R&D funds to the least developed regions to maintain economic cohesion.
JEL classification: O19, O38, R11.
Key words: Spatial Distribution, EU Technology Policy, economic cohesion, Convergence, Technology Indicators.
The European Union announced in Lisbon 2000 the objective of becoming by 2010 the most competitive knowledge-based economy in the world, and committed itself to undertake all necessary reforms in national and Community policies to achieve this goal. This strategy was based on the firm conviction that government policy can positively affect the long-run growth rate of the economy through economic incentives for the accumulation of various forms of capital and through the promotion of technological innovations. Such a conviction relies on the postulates of endogenous growth models (Romer, 1986 and 1990; Lucas 1988) and has motivated the proliferation of numerous national and European technology programs over the last decades. The idea behind each of these programs is the following: R&D generates innovation and new technologies, and innovation and new technologies generate then economic growth. This will happen because new technologies increase the productivity of production factors and therefore have a positive supply side effect on the growth potential of the economy. If this linear R&D—Tech/Innovation—Growth mechanism holds, then economic policy authorities would be very interested in promoting innovation and technology through strong R&D programs in the first place. Nevertheless, the relative success of these programs in achieving real innovation, and the relative success of these inventions in effectively generating higher rates of economic growth is still a matter of debate. The question of whether technology policies have really had any significant role in promoting economic growth or improving economic cohesion, still needs to be answered. Note, however, that the resolution of such research question would imply the development of a qualitative study based on the description of different policy initiatives which would complicate enormously the attribution of causality relationships between technology policies and economic performance. Instead, a better strategy is to study the evolution of some important technology indicators (mainly R&D spending and patent applications), assuming that there exists a connection between technology policies and technology outcomes in terms of R&D spending and patent applications. By doing this, a systematic quantitative analysis can be developed. The research question could then be reformulated as follows: Have R&D spending and patent applications had any positive or negative effect on economic growth and cohesion? This is the question that this paper will answer, and in doing so, the article not only wants to contribute to the debate on technology and growth, but also wants to investigate the possible existence of a trade off between economic growth and economic cohesion mediated by technology policies in general, and by technology indicators in particular. Aware of the likely existence of this trade off, Community policies have combined until now economic growth initiatives —such as R&D and technology programs— and explicit actions for economic cohesion —mainly through the structural funds— (Peterson and Sharp, 1998 and Pavitt, 1998). Now that these policies are being questioned in the current debate for the reorganization of European funds and policies it is crucial to link the answer of the research question that motivates this paper to the possible existence of the mentioned trade off. In order to do this, section 2 studies the spatial distribution of technology indicators over the last decade. Since the main finding of this section is that regional government R&D spending has converged while total R&D spending has diverged over the last decade, section 3 and section 4 focus on the likely different effect that these two R&D indicators may have had on economic performance. Therefore, section 3 re-interprets the evolution of these technology indicators vis á vis economic cohesion, and section 4 replicates the analysis for economic growth. Finally section 5 recapitulates and concludes.
2. Spatial distribution of technology indicators over the last decade
Technology policies are very difficult to measure quantitatively, and therefore their analysis has to rely on a set of technology indicators that approximate different phases of these policies, assuming that they follow a certain input-output sequence. Following the trend in the specialized literature we use total R&D expenditures by all sectors in % of GDP (TERD) as the best technology input indicator. The idea that total expenditures in R&D is a good indicator of technological innovation is basically derived from the so-called linear model of innovation, which assumes that investment in basic research is strongly positively correlated with technological innovation in the market place. Independently of whether this assumption holds or not, this is the best indicator to have an idea of the resource allocation to R&D in a particular region.
As an indicator of technology output we take the number of patent applications per million people. This is the so-called «inventiveness coefficient» and should be interpreted with care since Southern European regions are much less inclined to fill in patents for innovative products of processes (European Commission, 1997: 349). In spite of this fact, this is the best indicator to give an idea of the technology output intensity in a particular region . Finally, because we want to analyze separately if publicly finance policies have a different relative impact than the previous standard technology indicators, we analyze separately the evolution of government R&D expenditures (GERD), which is in itself a portion of the more general total R&D spending by all sectors . The use that we make in this section of all these indicators is twofold: first we just describe their spatial and temporal evolution, and then we report the results of a systematic convergence analysis whereby the common measures of economic and technological convergence are calculated and reported. In this respect, although in the specialized literature there is an open debate on the relative merits of different convergence measures , the two most popular measures are: the beta-convergence and the sigma-convergence. The former implies that the poor countries (regions) grow faster than the richer ones and it is generally tested by regressing the growth in per capita GDP on its initial level for a given cross-section of countries (regions). In turn, this beta-convergence covers two types of convergence: absolute and conditional (on a factor or a set of factors in addition to the initial level of per capita GDP). Under sigma-convergence we mean the reduction of per capita GDP dispersion within a sample of countries (regions) (see Barro and Sala-i-Martin (1995:11) for further details). We begin with the simplest indicator of all: the absolute beta-convergence index. Xie, Zou, and Davoodi (1999) elaborate a endogenous growth model based on Barro (1990) and Devarajan, Swaroop, and Zou (1996), where the production function has private capital and different components of government spending. Assuming a Cobb-Douglas production function, these authors obtain that the growth-maximizing share of a component of government expenditure in total government expenditures is equal to its elasticity divided by the sum of elasticities of all the components. Following this model, Sanz and Velazquez (2004) show that if output elasticities with respect to each component of government expenditure are similar across countries and that governments maximize growth, we should expect convergence in the composition of government expenditures among countries. Thus, as long as the elasticity of growth with respect to public R&D spending is similar across countries, we should expect convergence across public R&D spending in EU countries. Indeed, Gemmell and Kneller (2002) show that long-run growth elasticity of productive expenditures exhibit a high degree of uniformity across OECD countries. We start with the examination of β convergence, with the object of evaluating whether regions that have a higher public R&D spending increase (decrease) this percentage to a lesser (greater) extent than regions in which public R&D spending is lower. We will also adapt this analysis to aggregate R&D spending and patents in order to evaluate whether regions in which aggregate R&D spending and patents are lower have higher rates of growth. In this way we will be able to compare convergence in public R&D spending with aggregate R&D spending and patents. For this purpose we use the well-known ‘Barro type regression’: In (TIit) – In (TIi,t–1) = αi + β In (TIi,t–1) + εit 
where: TIt : is the Technology Indicator (patents, R&D, etc.) in year t. i: 205 regions of the EU at the NUTS II level of disaggregation for regional convergence t: all the years in the period 1989-2000 α: regional dummy. β: coefficient reflecting the existence and the speed of convergence. According to this equation, if the coefficient β takes a negative and significant value, there has been a convergence process in this technology indicator. Also, there would be absolute convergence in two cases: firstly if the GLS estimator is unbiased and hence we do not include any other variable apart from the previous year’s value as an explanatory variable for the change of rate; and secondly if only the within estimator is unbiased, but we can not reject the hypothesis of country dummies being equal for all the regions (De la Fuente, 2000). In this case all the regions will converge to the same steady state. Then, because the existence of beta-convergence is a necessary but not sufficient condition for convergence (Barro and Sala-i-Martin, 1992), we also compute the standard deviation of the logarithm of each technology indicator. In this context, the sigma-convergence explores if the dispersion among the different measures of technology inputs or outputs among European regions has been reduced. Finally, to complement and illustrate the results provided by the beta-convergence and sigmaconvergence analyses, we also plot Tukey’s box-and whisker plots for all technology indicators under study. The Tukey’s box-and-whisker plot is a histogram-like method of displaying data, where the box ends at the quartiles Q1 and Q3, and the statistical median is represented by a line that crosses the box. The farthest points that are not outliers (i.e. that are within 3/2 times the interquartile range of Q1 and Q3) are connected to the box by the «whiskers», and for every point that is more than 3/2 further away the end of the box, we draw a dot. To put the previous spatial distribution of total R&D expenditures in context, it is very important to note that statistics at the regional level show the more severe disparities between regions that remain hidden in statistics at national level. This is specially true for data on technology indicators. For example, disparities in technology input (TERD/GDP) and output (patents per million people) are as much as 20 and 55 times respectively higher at regional than at the national level. If one looks at the distribution of European regions that invested most in total R&D (as % of GDP) in 2000, we observe important disparities. Among the regions that invested most we find Braunschweig (7.19%), Stuttgart (4.92%), Oberbayern (4.79%), Pohjois-Suomi (4.73%); Pohjois Suomi (4.73%), Uusimaa (4.09%) and Tübingen (4.31%). And among the regions that invested least we find Calabria (0.24%), Castilla-la Mancha (0.22%), Sterea (0.18%), Dykiti Makedonia (0.07%) and Notio Aigaio (0.06%). In 2000, the EU’s average regional R&D spending was 1.22% of GDP with a 1.01 standard deviation. The spatial distribution of patent applications presents more disparities across regions than the distribution of total R&D expenditures. There are many regions which in 2000 filled out less than 4 patent applications per million people. Among them we find for example, Dessau (3.3), Andalusia (2.8), Molise (1.9), Galicia (1.5) or Calabria (0.9). On the opposite side, there were many regions which filled out more than 180 applications per million people. These were the cases of Koln (189.3), Berkshire (197.1), Stockholm (219.7), Noord-Brabant (266.8), Darmstadt (306.6) and Oberbayern (441.95), among others. Such a degree of disparity placed the EU’s average of patent applications per million people at regional level in 152.8, with a standard deviation of 147.9 in year 2000. It is worth noting that once controlled for the outliers, the regional disparity in technological development is not so high. This is because patenting activity is Europe is dominated by a small set of regions (an «Archipielago» of ten regions as suggested by Hilpert (1992)), with all others making only a marginal contribution. When compared to the spatial distributions of the two previous technology indicators, the regional distribution of public R&D (as % of GDP) is less dispersed. While there is a group of regions with very low levels of public R&D spending that range between 0.01 and the 0.04 of the region’s GDP (Schwaben, Sterea Ellada, Oberfranken, Koblenz, Rioja and Voralberg), there is another group that spends more public funds in R&D but at a moderate distance (Berlin 1.1%, MidiPyrénées 1.46% and Flevoland 1.87%). In 2000 the average level of EU’s regional public R&D spending remained at 0.19% of GDP with a standard deviation of 0.27. In view of the situation that technology indicators presented by the end of 2000, the question is now whether the spatial distribution of these indicators converged, diverged or remained intact along the last decade. In first place, Arellano and Bover (1990) test show that there are significant individual effects (see tables 1-3). There has been conditional convergence in R&d figures and in patent applications. This overall convergence has been, however, stronger in terms of patent applications than in total R&D expenditures. As the different coefficients in tables 1-3 show, the same has occurred with government R&D expenditures which have converged at a higher speed than any other technology indicator. It should be pointed out, however, that the results obtained in column 2 of tables 1-3 may be biased by the ‘country effect’, i.e.: by the fact that technology is more affected by the development of the country to which regions belong than by the actual features of the region. Consequently, we proceed in two ways to confirm that there has been regional convergence. First, in equation (3) we estimate convergence for the 205 regions including a dummy for the 15 Member States that takes value 1 if the region belong to a particular country and 0 if otherwise. Thus, we reduce the spatial self-correlation caused by the fact of the regions belonging to the same geographical areas (Armstrong, 1995). In this way, we obtain the results very similar to column (2) in all the tables. Second, in equation (4) we estimate regional convergence but taking the regional technological indicators in relation to the country average to which each region belongs. By means of this procedure, similar to that used in Rodríguez-Pose (1996), a very similar estimate to column (2) and (3) is obtained. Hence, from both procedures, it may be verified that, apart from the ‘country effect’ there is a technological convergent tendency specific of the regions. Furthermore, results corroborate that convergence in public R&D spending is higher than in aggregate government R&D spending. In fact, the different path of convergence of total and government R&D over the 1990-2000 period intensified during the second half of the nineties up to a point where regional total R&D expenditures started to diverge. This progressive divergence between both measures of R&D spending probably reflects the impact of the rapid expansion of private R&D spending as a share of total R&D expenditures. During the second half of the nineties while private R&D investment boosted, public R&D expenditures remained frozen at constant levels under the influence of general framework of budget stability. The sigma-convergence analysis reports very similar results to those provided by the previous beta-convergence analysis with only one exception (see figure 1). While both the beta and sigma-convergence analyses point to a convergence in patent applications and public R&D expenditures, particularly strong between 1996 and 2000, the picture for the evolution of total R&D expenditures is more heterogeneous. Apparently there exists beta-convergence and sigma-divergence over time. The existence of beta convergence would mean that regions with lower shares of total R&D in 1990 have increased their R&D expenditures at higher rates than those regions which started at higher levels. At the same time, the existence of sigma-divergence would imply that the dispersion from the average share of total R&D spending has increased. Nevertheless, these two results are not incompatible, because the existence of beta–convergence is a necessary but not a sufficient condition for sigma-convergence (De la Fuente, 2000). Random shocks may have increased temporarily the dispersion of total R&D expenditures even in the presence of beta-convergence or regions may be approaching their steady state shares (conditional convergence) with higher dispersion than at the beginning of the period. In addition, evidence of beta-convergence may be reflecting Galton’s fallacy, i.e. the tendency for regions to regress towards the mean (Quah, 1993). Just by looking at figure 1 it is easy to arrive at a very interesting finding: at the beginning of the nineties the fact of measuring the technological gap using different indicators really made a difference. In 1990, the existing technological gap measured by the sigma in patent applications (1.6) was twice the technological gap if the indicator to be used was total R&D expenditures (0.8). In 2000 the technological gap that both indicators measure is much closer, since in that year the sigma for patent applications was 1.6 while the sigma for total R&D expenditures was 1.3. Finally, all the dynamic evolution of the different distributions under study that was described in previous paragraphs is confirmed again when the three Tukey’s box and whisker figures are plotted. average total R&D spending has increased along time, as well as its degree of dispersion. However, the average level of public R&D has remained almost constant along the past decade and so has its degree of dispersion (plot 2). Finally, the average number of the log of patent applications has increased slightly in the last decade, while its dispersion diminished specially in 1995 and again in 2000. It is interesting to analyze the shape of the different Tukey’s box plots because they offer some new information on the sources of the existing disparities in the distribution of each technology indicator. The fact that dots are above the upper whiskers in the plots for total and public R&D expenditures implies that most regional disparities in R&D expenditures originate in regions that clearly spend long above the regional average. On the contrary the problem with patent applications is exactly the opposite: there is a significant number of regions that fill in very few patent applications and are therefore way below the regional average. Interestingly enough, and as we will see in next section , income disparities seem to be somewhat in between and find their roots in the existence of both very rich and very poor regions. Summing up the results reported until now, the most striking finding that the convergence analysis has provided is the empirical evidence showing that public and total spending in R&D have followed different dynamics over the last decades. If this different evolution has been translated into a different impact on economic cohesion and growth is the subject of the two following sections.
3. Spatial distribution of technology indicators vis á vis Economic Cohesion This section turns now, therefore, to explore the relationship between technology policy and economic performance. Following a logic structure we focus first on the link between the spatial distribution of technology indicators and the spatial distribution of income across European regions (also known as regional economic cohesion). Before this analysis can proceed it is necessary to briefly describe the evolution of the distribution of regional income per capita during the same period. Since we have assumed along the paper that there exists economic cohesion when the regional dispersion in GDP per capita diminishes, we want to estimate the relative impact that changes in the regional dispersion of technology indicators have on the regional dispersion of income per capita. To do so, we estimate the following equation, where all dependent and independent variables are transformed into their sigmadispersion indexes.
SigmaINCOME = SigmaPATENTS + SigmaTERD + SigmaGERD + ε 
regions in all EU and among regions by country. The influence of public R&D spending on economic cohesion is weaker8 but works in a similar direction. This direct relationship between the dispersion in all technology indicators and the dispersion in income per capita can be re-interpreted in view of the actual evolution of each indicator along the nineties that was portrayed in section 2 of this paper . A 10% increase in the dispersion of total R&D expenditures as the one occurred between 1998 and 2000, produced a 0.3% increase in the dispersion of income distribution across regions in Europe. On the other hand, a 10% decrease in the dispersion of public R&D expenditures as the one occurred between 1993 and 1995 and again between 1997 and 1999 produced each time a reduction of 0.1% in the income dispersion across regions in Europe. Apparently, the experience of the nineties shows that while the distribution of total R&D spending became more unequal (led by an unequal expansion of its private component) the only reason why this did not turn into a more unequal distribution of income per capita was due to the compensating effect performed by public R&D spending and patent applications. As the distribution of patents and public R&D converged, income per capita converged too. Since the only indicator that can be directly affected by policy-makers is the share of public funds that they dedicate to R&D activities, it looks like the government R&D has been used purposefully and successfully along the nineties to reduce the economic disparities that other technology indicators promoted. Whether this «cohesive» role played by public R&D expenditures has had any damaging impact on the rate of economic growth of these regions is a question that remains for the final section.
4. Spatial distribution of technology indicators vis á vis Economic Growth
There is a long tradition of empirical and theoretical studies on the role that technology plays on economic growth. While some recent works have found that economic convergence depends on a set of factors among which technology is only one of them (Paci, 1997; Dunford and Smith, 2000; Tondl, 2001), others have emphasized the decisive role that technology plays for long run economic convergence (Fagerberg, Verspagen and Caniels, 1997; Paci and Usai, 2000; and Paci and Pigliaru, 2001)9 . In order to study the relationship between the three technology indicators and economic growth this final section proceeds as follows: first, we simply study the correlation between the three technology indicators and income per capita. Then, we present the results of a multiple regression for the impact of technology on economic growth (measured as the annual change in GDP per capita). The correlation analysis provides clear-cut findings. The same occurs with the correlations between income per capita and total R&D spending (0.4). On the contrary, the correlations between income per capita and public spending in R&D are much weaker (0.15) and dilute over time. The joint role that all those technology indicators have in explaining economic growth measured by the annual change in income per capita can be discerned by estimating the following equation10:
∆GDPit = α0 + βo (GDPli,t–1) + β1 (TIfi,t–1) + βg (TIgi *cohesiongi,t–1) + + δt + ωp + øi + εti 
where: ∆GDPit: is the annual change in income per capita β1: is the coefficient reflecting the effect that all technology indicators (patent applications, total R&D spending, and public R&D spending) have on regional economic growth. βg: is the coefficient reflecting the effect that all technology indicators (patent applications, total R&D spending, and public R&D spending) have in regional economic growth in cohesion countries (Spain, Greece, Ireland and Portugal). The reason for including an interaction for the group of cohesion countries is that the relation between technology and growth might be different for different clusters (Clarysse and Muldur, 1999: 4). Clearly cohesion countries share common initial conditions and in this respect they can be considered as a separate cluster. δt : is the time dummy; øi : is the region dummy; ωp: is the country dummy. Note that growth might influence R&D spending decisions of both the business sector and the government. Growth increases government revenues, which in turn would raise the resources allocated to R&D spending. Companies in countries recording high growth rates may also devote more funds to R&D activities. For this reason, we introduce the lag values of the independent variables. In fact, regression (3) resembles the Beck and Katz´s dynamic model (1995, 1996 and 2005). Indeed, estimating equation (3) yields the same results as estimating GDP on its lagged value and independent variables. Thus, we are capturing long-term relationship between GDP and patents, aggregate R&D spending and government R&D spending. Moreover, endogeinity is not an issue anymore, as long as there is not first order autocorrelation in the error term (Beck and Katz, 1996). In the absence of first order autocorrelation, the error term will not be correlated with independent variables. Furthermore, Beck and Katz (1996) claim that «If the errors show serial correlation in the presence of a lagged dependent variable, the standard estimation strategy is instrumental variables. While this has fine asymptotic properties, it may perform very poorly in practical research situations (...) Thus it may well be the case that it is better to estimate with OLS, even in the presence of a small, but statistically significant, level of residual serial correlation of the errors». This is our case, The Lagrange Multiplier test shows that lagged residuals are marginally significant in predicting residuals from equation (3). Therefore, equation (3) is estimated through GLS with robust standard errors. Results are reported in table. As can be observed, real convergence is once again confirmed: the lower the existing regional income per capita in t-1, the higher the subsequent economic growth. In addition, the contribution of patent applications in t-1 to subsequent economic growth in year t is very positive. An increase of 1% in patent applications produces an increase in regional economic growth of 0.017. However, this effect is not so strong for cohesion countries. This can be interpreted as follows: where the stock of patents is very low, one additional patent is not sufficient to start economic growth. Instead, the innovative effort required to produce an isolated patent could diverting resources from more productive activities. More importantly, the role of total R&D expenditures is also strongly positive for all countries (including cohesion ones). The crucial impact of total R&D for economic growth is somewhat at odds with the insignificant effect that public R&D spending has on economic performance. However, this weak (even negative) short run impact of public R&D on economic growth, turns into a very strong and positive influence in the medium run. An increase of 1% in public R&D spending today is likely to increase the rate of growth by 0.04% in four years from now12. This 4-years lagged positive effect of public R&D on economic growth holds also for cohesion countries, what is very important given the fact that public R&D spending has to compensate for the low presence of private R&D initiatives in these regions.
Conclusion The study of the evolution that the distribution of regional technology indicators has experienced over the last decade has provided some clear and important findings which can be very useful to inform future economic policy debates in the EU. First of all, some technology indicators have converged among regions during the nineties (especially public R&D spending), and this has ran parallel to a real (though softer) convergence in income per capita levels. On the contrary total R&D expenditures have diverged across regions over time, as a result of an asymmetric expansion of private R&D activities during the second half of the nineties. Secondly, we have seen that total R&D spending increases economic growth, especially if this R&D activity is quickly transformed into new patent applications. Since innovation is the real key for economic growth, only when efficient total R&D allocations are easily transformed into new patent applications, economic performance improves. This positive effect on growth is not exclusive of total R&D expenditures, but also applies with a 4-year delay to public R&D initiatives. Finally and most importantly, in addition to this lagged positive effect on growth,government R&D spending has also demonstrated to be closely associated to regional economic cohesion in the short and medium run. When the dispersion of public R&D across regions diminished in the second half of the nineties, income disparities at regional level also decreased. Therefore, while technology policy based on pure excellence and efficiency criteria should remain as a policy tool for economic growth, this policy should be counterbalanced by European and regional policies which transfer funds to the least developed regions to maintain a minimum degree of economic cohesion. The results shown in this paper clearly demonstrate that if the current winds of reform succeed in curtailing the public financing of technology policies, the degree of regional polarization in the EU will most likely increase in the future.