Central Bank of Nigeria working paper finds forecast combination can improve short-term banking system liquidity forecasting

The Central Bank of Nigeria published a Working Paper that tests a modified forecast-combination approach for predicting Nigerian banking system liquidity by combining forecasts from eight driver-based models. The paper reports that combined point forecasts and midpoint interval forecasts track observed liquidity trends more closely than upper and lower interval bounds, indicating potential usefulness for short-term liquidity management, while noting the research does not represent Central Bank of Nigeria policy. The study uses monthly data from January 2008 to March 2024, with banking system liquidity measured as the average net liquidity position of the banking industry and the exchange rate represented by the NAFEM NGN/USD rate. Forecasts are generated from models linked to key liquidity drivers including total deposits, total bank loans, federal government expenditure, subnational FAAC allocations, the monetary policy rate, and standing deposit facility and standing lending facility volumes, then combined using a Weighted Least Squares framework with weights constrained to sum to one. In a four-period out-of-sample evaluation against an ARIMA (3,4) benchmark, RMSE results show the combined point forecast performed best at H=1 (33.56 versus 59.92 for ARIMA) and H=4 (66.24 versus 72.03), while ARIMA outperformed at H=2 and H=3. The authors recommend testing unequal weighting schemes that reflect relative in-sample predictability and expanding the set of liquidity drivers included in the forecast combination.