The role of oil prices in the forecasts of South African interest rates: A Bayesian approach*

- This paper considers whether the use of real oil price data can improve upon the forecasts of the interest rate in South Africa when using Bayesian vector autoregressive models.
- The full dataset includes quarterly measures of output, consumer prices, exchange rates, interest rates and oil prices (or its disaggregated positive and negative components), where the initial in-sample extends from 1979q1 to 1997q4.
- The evaluation makes use of rolling estimations and one- to eight-step ahead forecasts over the out-of-sample period 1998q1 to 2014q4.
- The results suggest that the model with the positive component of oil price tends to perform better than other models over the short to medium horizons.
- The model that includes both the positive and negative components of the oil price, provides superior forecasts at longer horizons, where the improvement is large enough to ensure that it is the best forecasting model on average.

This paper considers whether the use of real oil price data can improve upon the forecasts for the nominal interest rate in South Africa. We employ Bayesian vector autoregressive models that make use of various measures of oil prices and compare the forecasting results of these models with those that do not make use of this data. The real oil price data is also disaggregated into positive and negative components to establish whether this would improve upon the forecasting performance of the model. The full dataset includes quarterly measures of output, consumer prices, exchange rates, interest rates and oil prices, where the initial in-sample period extends from 1979q1 to 1997q4. We then perform recursive estimations and one- to eight-step ahead forecasts over the out-of-sample period 1998q1 to 2014q4. The results suggest that the models that include information relating to oil prices outperform the model that does not include this information, when comparing their out-of-sample properties. In addition, the model with the positive component of oil price tends to perform better than other models over the short to medium horizons. Then lastly, the model that includes both the positive and negative components of the oil price, provides superior forecasts over longer horizons, where the improvement is large enough to ensure that it is the best forecasting model on average. Hence, not only do real oil prices matter when forecasting interest rates, but the use of disaggregate oil price data may facilitate additional improvements.