Macroeconomic adjustment and the history of crises in open economies

Joshua Aizenman, Ilan Noy, 21 November 2012

Are countries that have previously experienced banking crises less vulnerable to them in the future? This column argues that, in fact, previous crises might make future crises more likely. There isn’t much evidence of a learning process from past mistakes because the regulator often lags behind banking’s pace of innovation. Preparing to prevent the last crisis does not prevent the next and it seems that the ‘too big to fail’ doctrine provides cover for banks, delaying the day of adjustment.

Looking at recent banking crises, Gourinchas and Obstfeld (2012) have identified domestic-credit booms and real currency appreciation as the most significant predictors of future banking crises in both advanced and emerging economies¹. An optimistic conjecture is that countries that previously experienced banking crises will tend to be more cautious. Efficient and fast adjustment of the public and the financial sectors to financial risks may reduce the probability of future banking crises. Yet, delayed adjustment of the public and the financial sectors – kicking the can down the road – may imply that past crises do not affect the probability of future crises.

Increased likelihood of banking crisis?

In Aizenman and Noy (2012), we examine the evidence and, intriguingly, failed to find efficient learning from past banking crises. A past occurrence of a banking crisis, on average, does not reduce and may even increase the probability of future crises². We also study the determinants of banking crises depth, finding that for middle-income countries, higher de jure capital account openness is associated with lower likelihood of a banking crisis, lower ratio of non-performing loans during the crisis, and higher levels of forgone output in the crisis’ aftermath. Yet, we find no evidence that the history of previous exposure of the banking sector to systemic crisis episodes seem to matter.

History and banking crises

We estimate the probability of occurrence of a banking crisis using a panel cross-country dataset for banking crises based on the newly updated banking crises database of Laeven and Valencia (2012). We use data for 1980-2010 for all countries for which the required data is available³. For control variables, we rely on previous recent research⁴. In addition, we add two controls. First, the history of banking crises for each country – a binary indicator of whether a banking crisis occurred in the previous decade – and, second, a similar measure of the history of currency crises⁵. Because the occurrence of the crisis will affect all of these variables, they are included in the specification with a one-year lag. Differences between high- and middle-income countries

A noteworthy observation is that once we include the data from the most recent years, the frequency of banking crises is not any different between high- and middle-income country samples, as shown in Figure 1. Historically, though, the middle-income countries were more exposed to banking crises. For the high-income sample (28 countries), the estimated model appears to be moderately useful in predicting banking crises. We find that a previous experience with banking crises increases the likelihood of another one occurring. For the middle-income country sample (74 countries), the banking history coefficient is positive and significant (though about half the size as for the high-income sample). In the middle-income sample, history-of-currency-crises variable also increases the likelihood a banking crisis (half as large as the banking crisis coefficient). We also find that higher banking liquidity decreases the likelihood of a banking crisis, while a pre-crisis credit expansion increases it⁶.

Figure 1. Descriptive statistics for panel dataset

Note: Standard deviations are in brackets, and the number of observations are in square brackets.

The current crisis

Focusing on 2008-2010, the estimated model provides strong predictive power. The banking-history variable is positive and significant, as with the longer time period sample, but with a larger coefficient. The same is true for the de facto exchange rate regime index, and our measure of credit expansion. What determines banking crisis magnitude? We also investigate the determinants of banking crises magnitude, relying on the newest version of the banking crisis dataset (Laeven and Valencia 2012), where the authors also include three variables measuring the depth of the crisis:

• The output loss, measured as deviations of GDP from a calculated trend.
• The fiscal costs, measured as a percentage of GDP.
• The peak non-performing loan (NPL) level reached, measured as % of total loans^{7 8}.

For our control variables, we rely on Hutchison et al. (2010) and Angkinand (2009).

Results

We find that higher GDP growth before the crisis is associated with fewer non-performing loans. Higher capital account openness de jure is associated with lower levels of NPLs and higher levels of economic disruptions, as measured by forgone output. For the fiscal cost proxy of the crisis, the only significant (and positive) coefficient is the IMF program participation indicator. We also find no evidence that the history of previous banking crises matters for the depth of the current crisis being experienced.

Regulators playing catch-up

A possible explanation for our failure to detect a learning process from past banking crises is that regulators and policymakers are learning, but at a speed that does not catch up with the dynamic evolution of modern banking. The regulator is frequently preparing to prevent the last crisis, and not the future one. as the contours of future vulnerabilities are fuzzy. In these circumstances, a possible remedy may call for slowing down the diffusion of financial innovations, treating them as risky until proven otherwise. This cautious attitude may call for more stringent leverage and reserve ratios, and blocking the introduction of financial innovations that may be ‘too clever by half’ for users and for regulators. Too big to fail?

A more troublesome interpretation is that too much political clout upheld by the financial system may cut the resources available to the regulator, and their ability to impose policies that are deemed too costly for and by the financial system. This concern arises especially in the context of the ‘too big to fail’ doctrine, where the private rents associated with excessive risk taking by the banking systems require adversarial relationships between the regulators and the private banking system. In these circumstances, policies aiming at curtailing the political clout of big financial institutions may help. Yet these policies may be hard to implement in countries dominated by few large financial players with cozy associations with the political process and with large conglomerates.

We interpret our results as consistent with a differential sectoral adjustment to crises hypothesis. The private sector, by virtue of its harder budget constraints, adjusts faster, whereas the government adjusts at a slower pace following a crisis. The financial sector may find itself in between the two. The ‘too big to fail’ doctrine associated with large banks provides them with a softer budget constraint, delaying the day of adjustment; for some, delaying bankruptcy. Occasionally, the separation between banks and the public sector is murky, further delaying necessary adjustments of the financial sector.

References

Aizenman, J and I Noy (2012), “Macroeconomic Adjustment and the Crisis History in Open Economies”, NBER Working paper,18527.

Angkinand, Apanard P (2009). "Banking regulation and the output cost of banking crises", Journal of International Financial Markets, Institutions and Money, 19(2), 240-257.

Demirgüç-Kunt, A and E Detragiache (2005), "Cross-Country Empirical Studies of Systemic Bank Distress: A Survey”, IMF Working Papers, 05(96).

Hutchison, Michael, Ilan Noy, and Lidan Wang (2010),”Fiscal and Monetary Policies and the Cost of Sudden Stops”, Journal of International Money and Finance, 29, 973-987.

Joyce, Joseph (2011), "Financial Globalization and Banking Crises in Emerging Markets", Open Economies Review, 22(5).

Laeven, Luc A and Fabian V Valencia,(2012), “Systemic Banking Crises Database: An Update”, IMF working paper, 12(163).

Laeven, Luc and Fabian Valencia (2010), “Resolution of Banking Crises: The Good, the Bad, and the Ugly”, IMF Working Paper, 10(146), June.

Noy, I (2004), "Financial Liberalization, Prudential Supervision and the Onset of Banking Crises - Empirical Findings”, Emerging Markets Review, 5(3), 341-359.

Reinhart, Carmen M and Kenneth S Rogoff (2009), This Time is Different: Eight Centuries of Financial Folly, Princeton, NJ, Princeton University Press.

Von Hagen, Jürgen & Tai-Kuang Ho (2007), "Money Market Pressure and the Determinants of Banking Crises", Journal of Money, Credit and Banking, 39(5), 1037-1066.

---

1 See also Reinhart and Rogoff (2009) and Laeven and Valencia (2010).

2 We are aware of the conventional wisdom that suggests that there are cases of countries where it is widely believed that a deep crisis caused changes that reduced exposure to future crises; Chile and Israel in the 1980s or Sweden in the 1990s are all frequently mentioned. Our empirical results imply that these cases appear to be the exception, and not the rule.

3 The Laeven and Valencia (2012) dataset details banking crises occurring after 1970, so we begin our sample a decade later in 1980.

4 In particular on Noy (2004), Demirgüç-Kunt and Detragiache (2005), von Hagen and Ho (2007), Joyce (2011), and Duttagupta and Cashin (2011). The list of independent variables we use is: per capita GDP, real GDP growth rate, a binary variable denoting hyperinflation (annual CPI>40), the de facto flexibility of the exchange rate regime, a measure of bank liquidity (deposit money bank assets as % of GDP), credit expansion (growth rate of deposit money bank assets), and the degree of openness of the capital account (the Chinn-Ito de jure index). Two other control variables were considered as they were included in some of the papers cited above, but both variables are not included in the specifications we present as they were never statistically significant: the real interest rate (nominal interest rate minus CPI) and the foreign exchange open position (foreign reserves/M2).

5 Since the two types of financial crises frequently occur together; this was dubbed ‘twin’ crisis by Kaminsky and Reinhart (1999).

6 The banking asset variables have the same sign as in the high income sample.

7 The fiscal costs are the bank restructuring costs defined as gross fiscal expenditures directed to the restructuring of the financial sector (e.g., recapitalisation costs). The dataset includes both the gross fiscal costs and the net costs (costs minus whatever the government managed to ‘claw’ back in the five years following the onset of the crisis). Since are interested in the depth of the crisis at the time it happens, we consequently focus on the gross fiscal costs instead of net costs. The NPL measure includes the peak measure of NPLs during the five years since the crisis onset (in cases where less than five years of post-onset data is available, the observed peak is recorded).

8 Surprisingly, the correlation between these measures of the magnitude of the banking crises is low. The fiscal measure is positively correlated with the two other measures (NPL and output loss), albeit moderately at 0.33-0.38. The NPL measure, however, has no statistically observable correlation with the output loss measure. It appears that all three measures are recording a different aspect of the depth of the crisis, and are not very closely related to each other.

Source