Forecasting the volume of mortgage loans with open Internet data in the period of noticeable changes in the Russian mortgage market

Abstract

Forecasting the volume of mortgage loans is important due to the necessity to correctly plan money flows and to ensure the liquidity of assets of credit organizations. At the time of the study, there are noticeable changes in the mortgage market in Russia associated with the coronavirus pandemics, the introduction of the preferential mortgage program, etc. In such conditions, flexible methods to make the forecast are specially required. Motivated by this, we develop an open Internet data-based approach for forecasting the volume of mortgage loans in the residential real estate market of Saint Petersburg (Russia). In particular, we first critically select sources of open Internet data about the volume of mortgage loans, real estate supply and demand, as well as data on relevant economic indicators. Furthermore, we analyze the selected data and use the most suitable ones as predictor variables in several statistical and artificial neural network-based models for forecasting the volume of mortgage loans (the target variable). It turns out that the usage of Yandex open data on the search query “buy an apartment in Saint Petersburg” (in Russian), the Russian Central Bank key rate and the dollar/ruble exchange rate data as predictor variable improves the forecasting quality. In particular, the ARIMAX model with the above-mentioned predictor variables outperforms all the models under consideration (Baseline, ARIMA, LSTM, MultiVariate LSTM, Transfer Learning) in terms of Mean Absolute Percentage Error (MAPE).