From forex forecasts to forex signals. Effect of the forecast move magnitude.

User Rating: / 0
PoorBest 
Written by Forex Automaton   
Friday, 29 January 2010 16:29

In the course of the trading system optimization, best returns have been seen to be obtained by ignoring forecast movesĀ  below a certain threshold, and acting on those above that threshold (that threshold was also called entry parameter). A small fraction of large returns has been seen to account for much of the positiveness of the Pearson correlation coefficient between actual logarithmic returns and their forecasts. Each Danica output contains forecasts for 14 forex rates, and a natural question is: how do I pick the ones to place trades? One might expect that the odds of success can be improved by selecting those markets where the next move is forecast to be large. I take version 0.5 of Danica forex system and study dependence of the correlation between the forecast and real logarithmic returns in day close on the relative strength of the forecast move.

I am going to stick to terminology used in physics and perhaps elsewhere and call an algorithm making a decision on whether and how to act on an incoming event in real time, a trigger. The term also applies to the signal communicating such a decision. Triggers vary in complexity involved in making the decision, level zero (L0) being the simplest. The higher the level, the fewer events pass through. Triggers use internal superficial characteristics of events to enrich the collected sample with events one is interested in. The concept can be carried over into algorithmic trading with almost no modification: ticks or regular updates on the market conditions, along with the candidate algorithmic trade ideas, are events, while trades are "accepted" events. If you will, it is the trigger that turns forex forecasts into forex signals.

The previos report introduced a way of classifying trade idea "events" according to the degree of consistency in the forecast direction of the candlestick's evolution (close, high, low). It demonstrated the advantages of the situation labeled "trigger word 3", where all three new components of a candlestick are expected to move in the same direction.

Product of forecast and acutal logarithmic returns for forex day close as a function of a ratio of a forecast move relative to the 10-day volatility

Fig.1. Product of forecast and acutal logarithmic returns for forex day close as a function of a ratio of a forecast move relative to the 10-day volatility of day close. In the inset: distribution of the above-mentioned ratio. Shaded areas mark one standard deviation around the points, calculated by comparing different markets, not different time intervals.

In Fig.1, the quality of forecast is gauged by summing products of logarithmic returns in daily close andĀ  the respective forecast. Successful forecasts result in positive products, unsuccessful ones -- in negative products. Moreover, correlation in magnitude of the forecast and actual moves drives the quantity along the vertical axis up, lack thereof -- down. The quantity plotted on the vertical axis is an average of such products within the classes of the forecast "events", defined by the bins of the horizontal axis. Despite large uncertainties associated with the choice of a market (market-to-market standard deviation of the quantity plotted), Fig.1 seems to show hints of the desired dependence. Namely, there is hope that one can bias the results higher along the vertical axis (meaning higher consistency of forecasts with reality) by focusing on large forecast moves. To some extent, the requirement of trigger word=3 already biases the distribution of forecasts towards higher magnitude: the hump in the inset plot of Fig.1 would not have been seen had not the condition of trigger word=3 been applied. Trigger word=3 also does the job of improving the odds of successful forecasts, as seen from Table 1.

market v0.5 v0.5+tw3 v0.5+tw3+move>0.8
AUD/USD 7.86 9.19 10.8
EUR/USD 2.19 3.04 5.71
GBP/USD 6.79 9.27 17.5
USD/CAD 2.30 2.95 5.9
USD/CHF -1.02 -1.35 -3.24
USD/JPY 3.33 3.62 1.35
AUD/JPY 3.98 5.45 12.1
CHF/JPY 2.82 4.31 3.95
EUR/AUD 15.1 9.99 14.9
EUR/CHF -0.410 1.25 -0.43
EUR/GBP 7.42 7.69 11.7
EUR/JPY 4.76 7.21 5.67
GBP/CHF 8.77 8.55 7.57
GBP/JPY 15.1 14.4 25.5
mean 5.64 6.11 8.50
standard deviation4.95 4.13 7.61

Table 1: Values of Pearson correlation coefficient, %, for various trigger conditions are compared to study effect of these triggers on forecasting quality for day close. v0.5: "plain" Danica; v0.5+tw3 -- as previous column, with an additional requirement of trigger word=3, v0.5+tw3+move>0.8 -- as previous column, with an additional requirement of a forecast move to volatility ratio above 0.8.

Fig.2. Data from Table 1 represented as a histogram.Entries correspond to individual exchange rate markets.

These estimates of market-to-market standard deviations differ considerably from those posted previously ([1], [2], [3], [4]). In the course of work on this report, a technical issue in the analysis software was found which resulted in the standard deviation (to be precise, RMS) estimates used in the reports linked above being essentially plausible-looking arbitrary numbers. The market-to-market standard deviation for close are about twice as high as previously reported, while those for daily high and low are several times better (lower) than previously reported.

Strictly speaking, with the present way of quantifying uncertainty -- that of inferring standard deviation by comparing runs of the system on different markets -- none of the options presented in Table 1 looks like a convincing evidence that the system is capable of predicting daily close. The plan is to look at temporal standard deviations characterizing consistency of forecasting quality within the markets in time. Should the results be significantly different (much lower temporal standard deviations), more individualized treatment of the separate markets within the system may be justified.

Bookmark with:

Deli.cio.us    Digg    reddit    Facebook    StumbleUpon    Newsvine
Last Updated ( Wednesday, 28 April 2010 16:47 )