Prediction Rupiah Currency Exchange against the USD with the Model Using TARCH (Threshold Autoregressive Conditional Heteroscedasticity)

Abstract

Economic time Series data often show a high volatility, for example the data currency exchange of rupiah against USD. Growth of the currency exchange before the economic crisis in 1990-1997 showed a relatively stable condition. However since the end of 1997, it shows a relatively unstable, an increasing and decreasing. In other words, the movement of data shows the high volatility. In econometrics it means that the variance of the residual time series data is not constant and change from one period to another period or contain heteroscedasticity elements. According to Engle, the changing residual variance is occurring because of residual variance not only influenced the independent variables but also depends on the residual in the past. Model that assumes the residual variants data is not constant in time series developed by Engle called Autoregressive Conditional Heteroscedasticity model (ARCH). Then ARCH model extended by Bollerslev GARCH model. But in many cases in the financial sector there is often a shock that is asymmetric (asymmetric shock). This means that sharp decline in the market (negative effects) will not necessarily followed by the increase in the market (positive effect) in the same size in another time. In other words, the negative effects are greater than the positive effect is called the TARCH model (Threshold Autoregressive Conditinal Heteroscedasticity). In the end of the task of writing test is not carried out asymmetric effects in the data currency exchange of rupiah against the USD, TARCH determine the best model for the rupiah currency exchange against USD and predict the rupiah against the USD with the best TARCH model. The used data currency exchange of rupiah against USD is from year 2003-2005. The results show that the existence of asymmetric effects in the data currency exchange of rupiah against the USD, using TARCH model, the best model is TARCH model (2,1), because results of the forecasting TARCH model (2,1) shows that the results forecasting not much different from the original data with Root Mean Squared Error 50,64878. So TARCH model (2,1) is well enough to apply to the data currency exchange of rupiah against USD.