Empirical determination of an appropriate earnings per share forecasting model for listed companies: a case of financial institutions in Kenya

Abstract

To date, little empirical work has been conducted in the area of technical analysis certainly not nearly as much as has been done on random walk at the NSE. What little has been done is neither conclusive nor are the results as consistent as those found in the random walk literature. Tests have been too simple, because they have utilized one tool at a time, rather than testing various models and reaching a consensus decision. This study was geared towards solving this problem.

The study presents the results of technical analysis attempts to forecast EPS for seven companies in the financial sector. Autocorrelation coefficients of the observed EPS series were utilized to search for patterns in the EPS series. Time plots and raw EPS data also played a major role in determination on non-stationary and pattern identification. Exponential smoothing procedures were also used to decompose the data to reveal unobserved components. It became clear after decomposition that there was an underlying trend in the data.

Exponential smoothing, AR (1), ARIMA (0,1,0) were fitted to the EPS data. Tests of model performance and accuracy were done using statistical measures. These summary statistics clearly showed that exponential smoothing is the most accurate model and is more superior than the simple naive methods of forecasting. We can therefore rely on forecasts from exponential smoothing method in making investments decisions.