Sampling and Forecasting Independent Data Via Clustered Bootstrap LSTM Models

In this study we propose a clustered bootstrap sample procedure which adds clustering to the existing bootstrapping techniques currently used in the independent data literature. To assess the improvement of the clustered bootstrap method, we compare it to the existing block bootstrap. We used different data generators to show the improvement introduced by clustering of the blocks and we further validated the practical utility of this method, we apply it to real-world financial data-specifically, the exchange rate between the South African Rand (ZAR) and the US Dollar (USD). The results of this study is measured with the use of the MSE, MAE, MAPE, RMSE and RMSPE. The results from the MSE, MAE and MAPE as well as the DM test all shows that the clustered bootstrap methods shows improvement in forecasting.