Judicial Indicators for Credit Markets in SSo Paulo

Abstract

This paper presents a new generation of judicial indicators for the credit markets proposed by Castro (2016), extracted from court diaries and websites in Sao Paulo, Brazil. The first generation was limited to counting of new, judicial cases pertaining to certain classes of disputes related to credit contracts, such as foreclosures of debt instruments and search and seizure in fiduciary liens, among other liabilities. Although that type of measure was significantly useful for tracking non-performing loans and to some extent, for improving macroeconomic forecasts, it suffered from a basic limitation, namely every contract under dispute weighs the same in the overall index. A natural extension of the measure is therefore, capturing unit values of the credits or claims under dispute, based on samples of cases on a monthly basis. In addition to unit values, we propose indicators based on other moments of the sample distribution of cases, such as the 10%, 50% and 90% percentiles, allowing for the distribution of judicial claims to change its shape along depending of the phase of the credit cycle. Another relevant improvement relatively to the first version is the significant increase of the sample period for the indicators: whereas the first version begun in October/2007, in this version dates back to September/2001. Besides computing aggregate, unit values of bad credits, the structure of the database allows slicing and dicing of the data to produce proxies of market conditions for specific segments of credit contracts, according to the nature of debtors (personal or corporate, small versus large firms) or creditors (banks versus non-banks) and the economic sector (industry, services, agriculture). Identification of the sector of firms is based on machine learning algorithms whereas other attributes are determined by regular expressions. The purpose of this paper is to evaluate the predictive ability of judicial indicators that are based on the case values of judicial claims, comparing forecasts with those obtained from baseline models that include traditional macroeconomic indicators. We conclude that the new indicators retain all the properties of the original indicators namely ease of computation and timeliness, confirming their potential as leading or coincident indicators of economic activity and moreover, their suitability for nowcasting.