Optimal Short-Term Inflation Rate Forecasting Model in Kenya: in Depth Analysis

Abstract

Listen

Translate article

Purpose: The Kenya economy experienced an increase in the price of basic commodities, increased unemployment rates and consequently reduced real wage levels due to job cuts. An accurate forecast of this unprecedented inflation rate could have cushioned the Kenyan population from its effects. Uncertainty over future inflation forecasting has caused detrimental and negative impact not only globally but also in the Kenyan economy. Combining several techniques of forecasting is an instinctual way to improve prediction performance as the limitations of one method are compensated by the strength of the other model. The general objective of the study was to develop an optimal model of forecasting short term inflation rate in Kenya. The specific objectives were to establish models of forecasting short term inflation rate using SARIMA, GARCH and hybrid SARIMA-GARCH family Models, select optimal model amongst the three models and Predict 12 months ahead inflation rate using the optimal model. Scope of the study from 2005 January to 2024 July. The study was anchored on monetary theory of inflation, Keynesian theory of inflation as well as rationale expectation theory of inflation. Methodology: Data was sourced from KNBS and CBK. The study was guided by positivism research philosophy. Explanatory research design was used in this study. Target population was 230 monthly observations. Sample and sampling techniques used was time series. PACF, ACF Dfuller, Kpss test, Philips perron test indicated the data was stationary after 1ST differencing. Findings: Statistic validation test results indicated SARIMA’s (1,1,1)(1,1,)12 adjusted 2 R was perfectly 1 indicating all variations in squared residuals were explained by lagged residuals. P- value for lagged squared residuals were significant at (0.00). F-statisticvalue was significant (0.020) suggesting overall model fit. Model stability test AR roots polynomial lied outside the unit circle. eGarch(1,1) model with lowest aic -0.534 was best in Garch Family models. Hybrid Sarima((1,1,1),(1,1,1) eGARCH(1,1) was identified. Comparison of Sarima(1,1,1)(1,1,1)12, eGARCH (1,1) and Hybrid Sarima((1,1,1),(1,1,1)12 eGARCH(1,1) using forecast accuracy revealed that hybrid model was the optimal model with lowest MAE 0.166 and RMSE 0.259. Diagnostic checks The Ljung- Box test (LL) and Q2 indicated non-significant autocorrelation p- value were greater than 05% indicating that the models residuals were white noise. The DOF/GED parameter (4.144466*, 1.101958, 6.977499) represented the degrees of freedom for the tdistribution and the coefficients where significant at 0.05% meaning the model’s assumptions for normal distributions were met. The coefficients of AR Normal 0.048140, T-Student −0.031854, GED −0.02396 and MA Normal 0.055167 ,T-Student 0.059645, GED 0.051390 terms were significant at 5%. Recommendations: The results implied that the model predicted a decrease in the Kenya’s inflation rate for the next 12 months. The study recommended that inflation rate would be hovering below an average rate of 10 within the next 12 months with high volatility up to July 2025 and policy makers should use this prediction for planning in order to maintain Kenya’s macroeconomic stability. Policy makers were also advised to use the hybrid model to forecast short term inflation rate 12 months ahead and in future years. The benefits of the study were added knowledge of hybridizing to researchers and to policy makers. This study therefore provided better inflation forecasts (Kenya) to be used for strategical planning for short -term effects of inflation by the government.