Introduction I start to be interested at Forex during my PhD thesis when I read with interest some old papers about Hidden Markov Processes in Economy. As part of the thesis I decided to investigated if it was possible to apply this methodology utilized in areas like speech recognition, Macroeconomy, Physics in the FX market .I found some good articles like Stylized facts of financial time series and hidden semi-Markov models written by Bulla. But the objective was to create a HMM model that outperforms simple technical indicators.So I start to built my own HMM model, I ask to Andrea Procaccini a funny and great Matlab programmer to cooperate to do something that could have interest also out of the University walls. Cyclex We call Cyclex the model investigated. I would try to explain with no mathematical formulas some of our results.Cyclex is an exchange rate forecasting model that outperforms a random walk at short horizons and appears to be robust and efficient over different sample spans. So an easy way to translate could be THE MODEL IT IS BETTER THAN TO FLIP A COIN AND DECIDE. Later we would try to answer to the questions How better is our model compared with the flip of a coin?How long we should trade to see some positive results?The main feature of the Cyclex theory is that statistically prices move in trends. The centerpiece of this forecasting model is represented by the generalization of the Engel and Hamilton's (1990) Markov-switching model and the work of Yuan, Forecasting Exchange Rates adding n multiple states in which trendless periods are considered in addition to the uptrend appreciation and downtrend depreciation regimes. To built a predictive Forex model we modified the standard Markov-switching model considering two important issues: Imposing only two regimes appreciation and depreciation is not consistent with the fact that almost all exchange rates occasionally exhibit range-bound behavior for a sustained period of time. The Hidden Markov Model is likely to overreact to irregular transitory blips in the data, you could imagine to model EUR/USD when Draghi, Carney or someone policy makers make some change in the Monetary Policy. Since financial time series, as FX price level, are often extremely noisy, the oversensitivity of the conventional models tends to induce instability in parameter estimation and misclassification of regime shifts, and in turn to undermine its forecastability.Cyclex is formulated to correct these two shortcomings. Forecasting In the context of exchange rates in the research sector , the standard measure of goodness, with the efficiency of a given predictive model, is obtained by comparison with the random walk. Probably a trader would not to be agree with this but try to be a compliant scientist have to quantify analitically.

The techniques generally used are:In Sample Forecast;Out of Sample Forecast.In Sample Forecast, (ISF) means using all the points in the dataset forming the time series to estimate the parameters of the model. Out of Sample Forecast (OOSF), instead,only makes predictions on points that have not been used to infer the values of the parameters that determine the model. It is common use the 80% of the observations to identify the model (set I of inference) and the remaining 20% (set F of forecast) to calculate the predictive power and this is approximatively what we did .
Results

Without trying to explain in detail the model we give the results.The following table contains the outcome that we obtained by using our model for the exchange rates of some of the major currency pairs.

So How much it is better ?

 Currency Pairs Gain AUD/USD 63.3% NZD/USD 25.6 % EUR/USD 58.0% GBP/USD 39.1% USD/CHF 38.8% USD/JPY 36.4% USD/CAD 23.9%
For example, in exchange AUD/USD, we observed an improvement of 63.3% compared to the models based on the Random Walk in the Mean Square Error relative to a daily time horizon.

How long do you should trade to have this results?

The data contain around 50 days of measurements.

Observations

Many traders probably make better profits but for the beginner like me could be great to know that there is a sort of model like this.
I am studing technical analysis system triyng to improve the model, there could be progress considering the introduction of trading techniques in Cyclex.
My friend and I probably are intention to traduce this model in a sort of trend indicator, any suggestion is welcome or if you have curiosity you could write throught Dukascopy messages at Durden.

Sorry for my english and good trade!
Translate to English Show original