Exploring the Interplay Between Early Warning Systems’ Usefulness and Basel III Regulation

Abstract

We analyse the ability of credit gap measures to predict banking crises by estimating the usefulness measure conditionally on policymaker’s preferences. The results show that the signals based on the credit gap indicators are most useful when the policymaker’s preferences regarding Type I and Type II errors are approximately equal. However, according to the current consensus, the preferences to avoid missing a crisis are higher than issuing a false signal. This means that the usefulness of the credit-gap-based early warning systems is likely to increase once the static Basel III regulative measures are implemented (assuming that their implementation results in lower financial crises’ costs).

Appendices

Appendix 1: Dataset

Appendix 2: Modelling the Effect of Changes in Capital Requirements on Financial Crises’ Severity

To assess the impact of the capital requirement introduction on the change in expected depth of recession or the severity of future crises, we use the model calibrated by Miles et al. (2013) to match historical experience dating back to almost 200 years.

The data are for the change in GDP per capita for a sample of 31 countries, and it starts, in some cases, in 1821 and lasts until 2008. The number of observations of annual GDP growth is almost 4500.

In line with Miles et al. (2013), we assume that the first difference of the log of per capita GDP (Y) follows a random walk with a drift and two random components. To capture capital requirement effect, we include an additional shock τ_t, which represents development banking insolvency as a response to the serious economic crisis. Like Miles et al. (2013), we assume that generalized falls in the value of bank assets are driven by changes in the level of incomes in the economy. Insolvency occurs when losses on bank assets exceed bank equity:

$$ \log \left({A}_t\right)=\log \left({A}_{t-1}\right)+\gamma +{u}_t+{v}_t+{\tau}_t, $$

(3)

where A_t—income (or GDP), γ—average productivity growth. u_t~N(0, σ²) represents the standard shocks in normal times. v_t represents a financial shock. It equals zero in normal times, but make take a very large negative value -b with small probability p and symmetric shocks of lesser magnitude ±c with probability q:

$$ {v}_t=\left\{\begin{array}{c}0, with\ probability\ \left(1-p-q\right);\\ {}-b, with\ probability\ p;\\ {}\begin{array}{c}+c, with\ probability\frac{q}{2};\\ {}-c, with\ probability\ q/2.\end{array}\end{array}\right. $$

(4)

The third shock τ_t represents the probability of an economic downturn becoming a full-scale systemic financial crisis. It links the value of capital adequacy ratio К and GDP losses. If banks have enough capital during a recession, the banking crisis does not occur (τ_t = 0), but it will happen otherwise. We implement this assumption as follows:

$$ {\tau}_t=\left\{\begin{array}{c}\delta \ast \left(\ \log \left({A}_{t-1}\right)-\log \left({A}_{t-2}\right)+K\right),\gamma +{u}_t+{v}_t+K<0;\\ {}0, oherwise.\end{array}\right. $$

(5)

We set К = 3% for the benchmark specification. Other parameters are reported in Table 2.

Table 2 Cross-section of countries

Under this parametrization, the model generates the distribution of GDP growth rates that is close to the empirical distribution reported by Miles et al. (2013). This comparison is reported in Tables 3 and 4.⁴

Table 3 Model parameters

Table 4 Statistics of artificial and empirical GDP growth rates

We proceed by conducting the following experiment. We change K from 3% to 10%, representing the increase in capital requirements in line with the Basel III recommendations. The new set of artificial GDP growth rates is computed, and several indicators of the severity of recessions in the alternative artificial datasets are compared.

The first indicator we calculate is the unconditional probability of observing a decline in GDP larger than a threshold P (we test P = 5% and P = 10%). The second indicator is the conditional probability of observing the decline larger than a threshold given that a recession takes place. The results are reported in Table 5. The estimates indicate that for P = 5%, the recession severity indicators are approximately halved when K is increased from 3% to 10%. The drop is even more significant if P = 10%. Arguably, these results may be regarded as a proxy for changes in the costs of a financial crisis under higher capital requirements. Accordingly, for the purpose of an early warning system’s usefulness evaluation exercise, we assume that the losses associated with the Type I error (i.e. missing a crisis) may be twice as low under Basel III’s capital requirements.

Table 5 Severity of recessions under different capital requirements