I work on time series data and I want to predict exchange rate volatility using different methods in order to compare them and choose the most efficient method
Most volatility modeling are handled by the Garch family estimates. So you can compare different catch type and see the that performs best...by checking the mae,Rmse and other criteria for checking forecasting accuracy
I do not think it is possible to forecast exchange rates based on data of the past. Of course you can use all sorts of models to calculate the trend and the volatility pattern. See for a more grounded explanation my contribution about the integrated exchange rate model on research gate. Published a feW years ago. I hope it will help you.
It depends. You can use time series models (These are commonly used). Therefore, you can use the criteria of goodness of performance of time series models like AIC and BIC. Generally, you are allowed to apply any stochastic processes meet the requirements (observations). In addition, it is possible to apply AI methods like neural networks.
Babak Jamshidi exactly what i would do in my thesis project; develop an exchange rate prediction model using neural networks combined with wavelet decomposition as a regression and decision support model. So I want to try conventional statistical methods to approve the improvement brought by artificial neural network models on the prediction of exchange rate volatility and their ability to adapt to abrupt changes.
As per your query, it seems that you have single series. In this case you can apply either symmetrical or asymmetrical GARCH. Symmetrical GARCH includes Standard GARCH (also known as plain vanilla GARCH) while asymmetrical GARCH consists of Exponential, GJR GARCH (also known as Threshold GARCH), APARCH, IGARCH etc. But you have to ensure that you must of volatility clustering and ARCH effect before applying these tests. Apart from this, you can check either the series has symmetrical effect or asymmetrical effect.
The use of ARCH and GARCH models have been applied but the results are not as per expectation. Exchange rate volatility is difficult because it reacts to so many variables. The in‐sample empirical results clearly show that the exchange rate volatility is best fitted to the GARCH‐MIDAS‐SLES model by including the short‐ and long‐term impacts of extreme shocks for all forecasting horizons. The out‐of‐sample results and robustness tests emphasize the significance of decomposing the effect of extreme shocks into short‐ and long‐term effects to improve the accuracy of the exchange rate volatility forecasts.
@ Chung Tin Fah, can you suggest a paper that employed GARCH-MIDAS-SLES?. Moreso, I believe Midas is used when we have variables of different frequency e.g monthly and weekly. @ Eya Bouzidi , Wavelet decomposition is good as you have noted, it allows you to check the variability of a series across different phases . I have also read some articles where the superiority of neuro fuzzy forecast is proven to be better than the traditional forecast method like Arima etc. Indeed its a good area to explore
Although there are a large number of methods to predict various phenomena with regard to time-series, an effective method to investigate might be neuro-fuzzy type2 methods. I have employed such a method in different problems and due to its impressive capability in handling uncertainties, the obtained results can be satisfying. However, there is an article, in which we have implemented a neuro-fuzzy type-2 coupled with different learning algorithms and might be helpful to take a look at it. If there is any question concerning the implementation of your architecture, I will be available to answer the questions.
Here are the article details:
Article Optimization and prediction of surface quality and cutting f...