Forex Automaton

This document explains typical output of Heidi -- an experimental free (payless but closed-source) ForexAutomaton hour-scale forex forecasting system or predictive model.

What data are published.

Raw output of the ForexAutomaton forecasting engine, automatically updated around (but not before) the beginning of every hour when at least one forex tick has been available during the preceding hour, is published along with sliding average figures of merit.

What this is not.

This system does not tell you how to trade on the basis of these forecasts. The output of the forecasting component ought to be considered as a first stage to a trade idea selection and money management system, yet to be created. The data presented are believed to be of non-negligible, but limited value and the user is expected to apply own critical judgement when evaluating the data presented here.

Selecting trade ideas and assigning capital properly is of particular importance for short-time-scale trading, since as the time scale shrinks, so does the expected profit per time step while broker's commissions or spreads (whatever applies) are the same regardless of your strategy.

Accuracy and figures of merit.

Figures of merit are updated and communicated with the same regularity as the forecasts themselves, and their present values are part of the forecasts. For times series with varying variance such as financial quotes, a simple difference between the predicted and real quote would be misleading.

Another idea, a precentage of cases when the system gave a right or wrong direction would not tell the whole story because a system may be right most of the time with negligible consequences, and is wrong rarely but in a disastrous way. What is important is not just the direction, but a measure of our confidence in it, or strength of the signal.

Our chosen figure of merit is Pearson correlation coefficient between the forecast and real logarithmic returns.

By construction, the Pearson correlation coefficient is a quantity bounded between -1 (the forecast move and reality are total opposites) and 1 (the forecast is perfect). A success or lack thereof for every hour makes a contribution to this quantity.

The coefficient is based on the product of the predicted and actual log returns. The product provides the desired incorporation of the signal strength, not just direction, into the figure of merit. A hypothetic rational operator of the system will not pursue small forecast moves, understanding this to be a noisy system. In order to make a large positive contribution, one needs a coincidence of a large move in a currency pair with a large forecast move in the same direction. This statement can be mirrored for a large negative contribution, while the rest of cases fall in between these two extremes.

Explanation of the hourly output.

Each daily forecast is an HTML document having a fixed structure, communicating certain pieces of information. Here is an explanation of these:

• System Name. A unique release of the trading system forecasting engine, coupled with an optimized set of strategy parameters is given a personal name. Names starting with a D denote systems with decision-making frequency once a day; names starting with an H -- once an hour. This experimental system launched in summer 2010 is called Heidi.

• Validity date. A point of time for which the forecast is given. Validity date is given in Eastern time, unless stated otherwise.

• Execution date. A point in time when the forecast was run. Note: back-testing forecasts for the chosen version of the system may be also archived on the Web. They contain appropriate disclaimer to distinguish them from "production" forecasts run in real time. For the back-testing forecasts, the execution date is a later date than the validity date.

• High, close, low (of the forex rate). In the table, these are arranged vertically in this order.

• Next: the essence of the forecast; forcast quantity characterizing the next day "candle" (day high, close or low), see Validity date above.

• Current: characterizes the "candle" for the hour which is just over, the 60-minute period ending just before the forecast is generated.

• Next-Current: the arithmetic difference between Next and Current. If it's positive, we have a "bullish" forecast; if negative -- bearish.

• Pearson(forecast,reality): moving average of the Pearson correlation coefficient between the forecast and actual logarithmic returns. Currently this is the main figure of merit for the portfolio-free trading system.

  By construction, this correlation coefficient is a quantity bounded between -1 (the forecast move and reality are total opposites) and 1 (the forecast is perfect). We report daily (24-hour) and 24-day (576-hour) moving averages and a "since inception" integral of this quantity. The latter one may include the back-testing period.

Bookmark with: