A high frequency indicator of credit in Peru: A Random Forests and dynamic network connectedness approach

Grandez R.; Pardo-Figueroa R. · Latin American Journal of Central Banking (2026) · DOI: 10.1016/j.latcb.2026.100204

Resumen

This paper contributes to the literature on early warning indicators for financial stability by developing a daily credit expansion indicator for Peru. We employ a two-step machine learning approach to achieve this. First, we use a Random Forest (RF) model to impute missing daily data within the weekly credit series. Second, we apply a Quantile Random Forest (QRF) model to identify periods of excessive credit growth by examining high-frequency financial variables such as credit spreads, interest rates, and stock market returns. Specifically, we focus on growth quantiles above the 95th percentile, which signal potential financial instability. This methodology integrates mixed-frequency analysis with machine learning to provide a robust early warning tool for detecting episodes of excessive credit growth in Peru. Our method's ability to manage nonlinear relationships and adapt to diverse economic scenarios offers substantial value to central bankers and macroprudential authorities focused on maintaining financial stability. Additionally, we analyze the financial connectivity among these series using a Quantile VAR (QVAR) model. To measure the transmission of shocks between these variables across different parts of the distribution, we employ the connectedness approach of Diebold and Yilmaz (2012) which is widely used to study spillovers in financial markets.