Abstract
Abstract : Predicting equity share prices could be useful to various stakeholders. The common methods used to forecast equity share price besides the naïve model are the Autoregressive Conditional Heteroskedasticity (ARCH) and General Autoregressive Conditional Heteroskedasticity (GARCH) models, however, no conclusion has been reached as to which model produces the most accurate predictions. In this research, ARCH and GARCH forecasting models (and their extended variants), as well as the Monte Carlo Simulation, were used to forecast price-weighted equity indices that were constructed from the South African, Nigerian, and Kenyan share markets. These three countries were selected based on their significance in the African continent due to the relative size of their economies and the liquidity of their share markets. The daily closing share prices for companies listed on the FTSE/JSE Top 40 Index, NSE Top 30 Index, and the NrSE Top 20 Index were collected between the 4th of January 2010 and the 30th of June 2015. The companies that were selected from each of these indices to construct the price-weighted indices for each country, were based on criteria to eliminate bias. Different autoregressive models were fitted for the mean equation. The EViews statistical programme was used to analyse the data. The ARCH effects were tested using the ARCH LM test. The ARCH/GARCH family models selected were GARCH (2,1), EGARCH (2,2), and EGARCH (2,1) for Nigeria, Kenya, and South Africa respectively. A Monte Carlo Simulation with 1 200 iterations was also performed to forecast the equity share prices. Post estimation and performance evaluation metrics were performed using the RMSE, MSE, MAD, and MAPE. The results based on the evaluation metrics indicated that the ARCH/GARCH models in-sample forecasts were more accurate than out-of-sample forecasts. The accuracy of the ARCH/GARCH models’ predictions was sounder than that of the Monte Carlo Simulation based on the evaluation metrics. Comparing the forecasting models to the actual graphs, in most cases the ARCH/GARCH models were closer to the actuals than the Monte Carlo II Simulation. The accuracy of the model predictions were also influenced by the sample size, the nature of the data, the leverage effect, and the macro economic conditions. In conclusion, the African equity markets cannot be predicted accurately using the ARCH/GARCH models and the Monte Carlo Simulation. The predictions from the forecasting models are not sufficiently accurate for investors, traders, and company management to use to make informed decisions. However, these predictions are better than the naïve model. The researcher also concluded that the markets are efficient, as the publicly available information cannot be used to gain abnormal returns. This study’s findings are similar to those of previous studies carried out in South Africa and globally.
M.Com. (Finance)