Behavioural early warning indicators of household sector vulnerability
Working paper 945
DOI:
https://doi.org/10.71587/x1q80491Keywords:
early warning system, Google Trends, Household behaviour, macroprudential surveilance, South AfricaAbstract
Conventional indicators of household financial vulnerability are published with substantial lags. This paper examines whether Google search behaviour can provide earlier signals in South Africa. Three behavioural early warning indicators are constructed from Google Trends data, capturing housing and mortgage demand, formal financial distress and informal credit substitution (mashonisa loans). Their predictive performance is evaluated using quasi-real-time recursive out-of-sample forecasts from January 2015 for seven banking, credit and housing outcomes at horizons of up to 24 months. Used individually, the indices achieve higher AUROCs than the credit-to-GDP gap and monthly macro indicators in six of the seven outcomes, with the clearest signals one to eight months ahead. The signal is statistically clear for household credit busts (Demand and Distress), deposit collapses (the informal-credit Shadow index) and house price booms (Distress), and lasts roughly 8 to 13 months. The indices also provide statistically significant incremental predictive information beyond established macro-financial indicators such as the credit-to-GDP gap, notably for household credit busts. The gains over the gap are robust to alternative event definitions, estimation choices and information timing, and disappear when the indices are shifted in time. Search behaviour therefore complements, rather than replaces, macro-financial data, and offers central banks a timely, low-cost addition to early warning systems, potentially helping them monitor household financial vulnerability where informal credit is hard to measure.
