Experiment

This section describes the experimental results performed with the aim to obtain a realistic trad­ers behaviour. Speaking broadly regarding the experimental design, the experiments were car­ried out using various time horizons of the year in order to confirm that the collective trading agents’ behaviour replicate, to a certain extent, the col­lective FX market traders’ behaviour.

Validation Design

To validate the accuracy of the trading agents’ behaviour in resembling the collective behaviour of real FX market traders, we define the stylized facts of real FX market traders behaviour by ob­serving their micro-behaviour in the market, as reported in Section 5. Following this, we observe the collective behaviour of the trading agents to establish their stylized facts. Using the identified stylized facts of the collective behaviour of real FX market traders and trading agents, we measure the similarity.

Results

In this section, we present the experimental results under which the stylized facts of the collective behaviour of real FX market traders are exhibited. It is worth mentioning that several experiments using different sets of the model’s parameters were carried out before obtaining a realistic traders’ behaviour in the market.

The validation of the agent-based models of traders is based on correlation and the seasonality of FX market trading activity. From the simulation outcome, we found a positive linear relationship between the numbers of buy and sell orders, re­flected in their correlation coefficient, which is 1. Additionally, the correlation coefficient is 1 for the number of opening and closing positions.

Figure 3a shows the intraday seasonality of EUR/USD trading activity from the simulation outcome. The patterns dynamic captures the double U-shape or ‘camel-shape’ pattern reported in the literature (Dacorogna, et al., 2001; Ito & Hashimoto, 2006). We can see that it follows the same pattern as the one presented in the intraday seasonality of the real FX market EUR/USD trading activity in Figure 2a, with two peaks, the second of which is higher than the first. The first trading activity peak takes place when the Tokyo and London trading sessions overlap around 08:00 GMT, while the second one takes place when the London and New York trading sessions overlap around 13:00 - 16:00 GMT. The minimum hourly activity occurs around 3:00 - 6:00 and 20:00 - 21:00 GMT. This is when Tokyo and Sydney markets are open, and the New York market is starting to close, respectively. Additionally, we can see that the trading activity starts to peak within the opening trading business time of the market centre in the morning. To illustrate, the number oftrading activity increases at 01:00 GMT which is when the Tokyo trading session opens, then when the London trading session opens, it increases again at 08:00 GMT. When the New York trading session starts, the number of trading activity increases sharply at 13:00 GMT, which is when the London and the New York markets’ trading sessions overlap due to the fact that more traders are participating in the market.

Figure 3b shows the intraweek seasonality of EUR/USD trading activity in terms of the simu­lation outcome. We can see that the seasonality dynamic of the trading activity exhibits similar behaviour to the intraweek seasonality of the real FX market in terms of EUR/USD trading activ­ity in Figure 2b. The number of trading activity declines sharply during the weekend due to the small number of participants. The highest num­ber of trading activity on each day occurs when London and New York trading sessions overlap around 13:00 - 16:00 GMT.

8.