A robust performance measure of trading strategies is one of the most important ingredients in the development of new models and also in their use. In Section 11.3, we discuss different possible performance measures and we, derive two risk-adjusted ones for investors with risk-averse preferences. Maximizing these measures is equivalent to maximizing the expected utility of an investor.

To construct successful trading strategies is not an easy task and many possible mistakes must be avoided during the different development phases of new models. We shall describe here some of the main traps in which new system designers generally fall and provide some ideas as to how to construct more robust trading strategies. In the following sections we will also give a short description of the various components needed in trading models and a specific approach using genetic algorithms to obtain more robust optimization results.

11.2 real-time trading strategies

In the assumption of a heterogeneous market, there is no trading strategy that is absolutely better than other ones. Which strategy to choose will depend on the trading and risk profile of the investor. This is confirmed by the existence of many different types of portfolio and investment strategies in the financial markets. It is also why we use in this study different trading model algorithms. We believe that these new investment strategies will simply contribute to, and not fundamentally change, the heterogeneous composition of financial markets.

To be useful, real-time trading models must provide realistic trading recommendations that the user can follow. This means that the models should do the following:

Give a warning a few minutes in advance of a deal.

Not change recommendations too rapidly.

Not give recommendations outside business hours.

Take into account market holidays.

Support stop-loss (around the clock).

In this section, we present the basic system architecture that we use in our real-time trading models and discuss the main components needed to transform available price quotes into actual trading recommendations. The model is divided in three main parts, that is,

Generation of the trading model recommendations.

Receipt of the simulated positions by the simulated trader.

Generation of the model statistics by the performance calculator.

Figure 11.1 depicts the overall structure and data flow of a simple real-time trading model. The next subsections describe these different components.

Filtered Quotes

Trading Model

Market - Time

Current Return Calculator

Current Return

Gearing Calculator

Performance Calculator

Historical Statistics

Model Statistics

Stop-loss Detector

Gearing

Gearing

Recommendation Maker (Deal Filter)

Simulated

Recommendation

Trader

Simulated Position

Display

TM Portfolio

Warning signal to users

Users

FIGURE 11.1 Data flow of prices and deal recommendations within a real-time trading model.

TABLE I I. I Market constraints.

FX market business time constraints for the trading models running for different geographical markets. The markets are listed roughly in the order of their opening times in GMT.

11.2.1 The Trading Model and Its Data-Processing Environment

As in forecasting or other applications, trading models depend heavily on the quality of the financial data that are fed to the program. Problems related to bad or invalid data can play an important role at different stages of the decision process. For instance, bad data can disturb the computation of the model indicators and then imply a partial or complete loss of the prediction power related to these indicators. One other sensitive part is the computation of the current return of the open positions. The current return is often used to trigger stop-profit algorithms, or exit specific positions. Any invalid price that passes through the filter incorrectly can produce a long-term perturbation effect on the trading abilities of the system, especially if it is used as a transaction price. To avoid data-related problems, a good trading system must include a special filter to cancel or postpone recommendations until a realistic transaction price is selected.

Trading Hours and Market Holidays Although some markets like the FX market operate continuously, individual traders or institutions generally partake of this market only for a portion of each day. Our models accommodate such users by incorporating the notion of business hours and holidays. Every trading model is associated with a local market that is identified with a corresponding geographical region. In turn, this is associated with generally accepted office hours and public holidays. The local market is defined to be open at any time during office hours provided that the trading model does not operate on a weekend or a public holiday. Typical opening hours for a model are between 9:00 and 17:00 local time, the exact times depending on the particular local market and traded instruments. In the case of FX, Table 11.1 presents typical opening hours of different geographical markets. Except for closing an open position if the price hits a stop loss limit (described in section 11.2.1), a model may not deal outside of the markets opening hours.