Does Money Matter for Predicting Overall Prices in Albania? An Analysis with Recurrent Neural Network

The purpose of this article is to assess the information content of monetary aggregates in predicting overall prices in Albania. The relevance of money is evaluated by comparing forecasts derived from no-money versus money-based models. Rather than employing traditional econometric models, an important innovation in our analysis is to use the Long Short-Term Memory (LSTM) technique of recurrent neural networks. These powerful tools allow higher flexibility than conventional functional forms for achieving the desired degree of forecast accuracy. After estimating the neural network parameters on quarterly data from 1993 to 2016, the forecast performance is then evaluated in a pseudo-out-of-sample exercise for horizons varying from one to twelve quarters ahead during the period 2017-2022. Preliminary results indicate that narrow monetary aggregates, particularly base money that is controlled by the central bank, have an important role in predicting prices at all horizons up to around two years. Contrary to expectations, the contribution of broader monetary aggregates M2 and M3 is found unstable across time horizons. The LSTM model results also uncover time-varying effects of monetary aggregates. We find evidence that the impact of money growth on overall price developments was weaker in the years before the pandemic, and it increased considerably during the accelerating inflation in the post-coronavirus and energy shock period. As it is the more recent period that matters for monetary policy, it is argued that money matters in the case of Albania (at least in particular economic circumstances) and due diligence should be dedicated to money-based price models.