Walk forward test

Q: We all know that future performance does not resemble past performance. If I have found a good trading model for a stock, should I perform some walk forward test? If yes, how can I reserve some price data (which has never been seen by the model) for the walk forward test? Technically, how can I conduct the test?
A: Our models are purely statistical. We took every care to avoid any chance of using forward info or any other technique, which could compromise optimization results in live trading. One could expect that the models will perform in future approximately same as they did in the past. At least as far as fundamental statistics will not completely change for given symbol.

Best model selection

Q: I am not sure about the proper means to assess the quality of a trading model (algorithm plus trading strategy) for a given stock. Is it purely based on the Performance Report? So, the model giving the highest % of winning trades and highest net profit should be the best? However, in most cases, the Expert Advisor does not give the best trading model in terms of performance report – why is that so?
A: Expert advisor selects the model with the best net profit. It does not relate to the number of winning trades. For instance, the strategy may well have the number of loosing trades exceeding the number of winning trades, while the average profit per winning trade will increase the average loss of loosing trades. Such a strategy will still be profitable. Therefore, we consider net profitability as the only reliable measure of efficiency of a strategy. It is combined, of course, with special statistics estimating stability of the results.
Best model selection

Predictors and indicators

Q: Some algorithms in user guide are referred as predictors, others – as indictors. What is the difference between these two kinds of algorithms in trading usage?
Price action
A: We offer classical indicator style algorithms intended as the analysis tool for historical data and as an auxiliary supplement in building custom strategies by the user. Indicators are very similar in their usage to the standard technical indicators common in most trading platforms.
Predictors, even though very similar in display to indicators, are much more elaborate algorithms in mathematical sense. They are designed not to simply monitor the price dynamics, but to predict future dynamics depending on existing historical prices. Predicted future price serves as the advice to trader to make immediate decision on the current positions to maximize returns. Trader may wish to perform certain actions depending on the predicted price change. For instance, trader may wish to close position, if he will see that the price is likely to move down on the next bar, according to predictor. In this way, predictors serve as precise trading advisors.

Recurrence of results

Q: We run your system several times on the same or slightly changed input data and obtain the results, which do not match exactly in each case. Does it mean that the results are unstable and not trustworthy?
A: Our expert system contains very complex forecasting algorithms. It includes wavelet regressions, neural networks, complex statistical optimizers and nonlinear filters. These technologies involve calculations thousands times exceeding in volumes all typical indicators supplied with trading platforms. All computers operate on finite precision arithmetic. It typically has 15 digits accuracy. In very long and complex calculations even small discrepancies tend to sum up and bring noticeable divergences preventing the results of several runs to coincide exactly even on the same data supplied. What matters here, is the tendency in behavior of such errors.
If the results of several consecutive runs do not differ considerably and has the tendency to match exactly with the increase in the input data volume, then we can tell that algorithm is stable and has the convergence point in the multidimensional phase space. In other words, we can say that algorithm has attractor, which guarantees statistical accuracy and consistency of the results on the ensemble of representative input statistics. Convergence is governed by Lyapunov exponents and other chaos measures. This style of convergence escapes full literacy and exhibits sporadic misfits and other phenomena. Key indicator is then steady decrease in the number and amplitude of such deviations with increase of data series length, which indicates underlying model convergence to stable result in terms of its phase space.
Results of all our algorithms generally coincide in consecutive runs. Coincidence is, however, not literal because of convergence aspects of inner optimizers. On short data series, coincidence is poor or even completely absent due to statistically insignificant data. Coincidence much improves on longer series and gets better with additional data added in most cases. It indicates correct model structure and its applicability to market forecasting. In many cases, good reproduction of results requires quite long data series of thousand points and more. Exactly this statistical convergence gives our technology fundamental benefits over the most common technical indicators.

Optimal market depth

Q: User manual gives special table of allowed input lengths for each algorithm. Why different algorithms require different length of input data? How we can choose this length optimally in each case to achieve the best performance?
Optimal market depth
A: We offer algorithms of different complexity levels. Simple algorithms typically work well on very short input data. However, many algorithms are adaptive. It means that they automatically adjust their settings based on input data to achieve best forecasting performance.
As with humans, algorithms have only one way to learn – they learn on the previous experience. This experience is passed to algorithm in a form of market history. The longer history is provided, the better experience will have algorithm to learn. As with humans, there is no guarantee that algorithm will learn well, provided it has good initial data. However, not giving enough data is solid guarantee that even good algorithms will have no experience to educate on it properly.
Each algorithm has the minimal survival data length. Without minimal data, it simply cannot live and operate. All initial data points on chart will be just void until this minimal limit is reached, indicating that algorithm is not operational until enough data history has been collected. However, it means exactly the survival limit. Calculation starts immediately as soon as the limit is reached. Nevertheless, it does not mean that calculation on these limit condions is perfectly accurate. Algorithm with the minimal data is as the newborn baby: it is very unnatural to expect from it to be market genius. You must teach it with enough historical data to expect better professional expertise. The more data you provide the better results you can expect.
On another side, the calculation time grows with the amount of data. On the real time stream forecast will appear useless, if it arrives later than the forecasted event really happens. Hence, the user must choose the balance between the desired accuracy and the allowed data depth to get the timely market advises depending on performance of computer.

Lag correlation analysis

StockFusion Studio includes the powerful lag correlation matrix for discovery of symbol dependencies. It calculates cross correlation of selected symbol with other symbols over a range of time lags. If strong correlation exists, then correlated symbol can serve as a sort of predictor.

For example, we see that EKK for some reason correlates with EVO on lag 16. This means that if EVO has some price change, it is 42% likely that EKK will have very same price change in 16 trading days after that. Similar dependencies should exist intraday also.

Portfolio correlation matrix

Portfolio correlation matrix is the powerful tool for your portfolio optimization. It allows optimal portfolio balancing be excluding from it closely correlated symbols. If you have several closely correlated symbols in portfolio, they similarly behave on market changes and do not add any stability to your portfolio. You must try to keep in portfolio only poorly related symbols. This generally increases portfolio stability because it will increase probability of some symbols growing when the others are on down slump.

StockFusion Studio gives you powerful option of optimizing your portfolio through correlation matrix analysis, which will highlight optimal combination of symbols in your portfolio.