Exchange rates are very important for business, banks, investors and of course small traders like us who want to profit from their volatility. If you are dealing with EURUSD, you will want to know whether to buy or dump EURUSD this week? One method used by us traders is to look at the charts and do some technical analysis. But there is a huge literature on how to predict currency exchange rates using statistical models.
Central banks need to model these exchange rates precisely in order to determine their monetary policy stimulus. You should read this FED report Insights from the Federal Reserve Bank of Dallas, it is explained why it is very difficult to predict exchange rates using econometric models. The author argues that fundamental data like inflation, interest rates, GDP, trade balances, money supply and other economic indicators are simply not able to explain the short to medium term currency exchange rate fluctuations.
Economic data like national income, trade balance, money supply, inflation, interest rates etc are far less volatile as compared to the exchange rates. On the other hand simple random walk model can predict the short term and medium term exchange rate fluctuations much better as compared to the econometric models based on the above fundamentals.
But don’t think the random walk model can make good predictions. It is just that it’s predictions are better than the predictions made using econometric models based on fundamental analysis. But not too good to make you think that this model can make good predictions.
Exchange rates depend on the market expectations. Supply and demand just like any other market is of course a key determinant of these exchange rates. There are several statistical models that have been developed to predict currency exchange rates. Some of these models are GARCH, Recurrent Neural Networks, Regression Neural Networks, Kernel based regression models etc. But almost all of them are not very good.GARCH models were the first one that were developed. Now these models are somewhat good on the weekly and daily timeframes. But on the intraday timeframes these models are not very good.
So if you have been thinking of using some neural net software to predict these rates than don’t think that the predictions are going to make you good buy and sell decisions. The only think that still works is technical analysis. Naked trading is what works. Naked trading means trading only with price action. Price action on the charts tells you what the buyers and sellers are thinking at that particular point of time.