Start AI trading with MyMate! | Vol.8 "Markets that AI agents are good at, markets they struggle with"

Markets it excels in and markets it struggles with

Many people may wonder in which markets the AI Agent excels and where it struggles.

This time, for each currency pair, USD/JPYUSD/JPY,GBP/USD,EUR/USD, we compare the AI Agent’srealized P/Landunrealized P/Lwithactual ratesto seewhat kinds of market conditions it handles well and how its performance changes around sharp market shiftswill be examined.

The data is the averages of realized P/L and unrealized P/L for all AI agents on each currency pair from September 1, 2020 to August 31, 2021, with the corresponding daily closing prices lined up.

USD/JPY

The table below compares the realized and unrealized P/L of USD/JPY AI with its rate.

Looking at the USD/JPY rate during this period, an uptrend began in early March, then a temporary decline, and by late June it formed a gradual uptrend with fluctuations.

Andthe period when USD/JPY AI performed relatively well was also from March to late June 2021.There are days with losses, but more days with greater profits, so it can be considered favorable.

On the other hand, around mid-September 2020 and around November 24, when the market fell, both realized and unrealized P/L shifted strongly positive; however, on other market declines, both realized and unrealized P/L were mostly negative, and positive P/L tended to occur during market uptrends.

In other words,USD/JPY AI tends to be strong during rising markets.

GBP/USD

Next, let’s look at GBP/USD AI data.

The most interesting part is the relationship between GBP/USD AI’s performance and the market from around November 2020 to late February 2021.
In the USD/JPY AI, trend markets yielded good results. In contrast, GBP/USD AI shows unrealized P/L positive but realized P/L remaining negative.

So, in what kind of markets does it generate profits?
When looking at spots with large positive shifts in realized and unrealized P/L, around mid-September 2020 during a market drop and around May 12, 2021 during a market rise, the tendency to be positive varies with time.

From these,it can be imagined that the AI learns continuously to decide when the trend has shifted most recently and when to close positions.
For example, around December 8, 2020 and around July 21, 2021, price movements are similar, but the former shows large negative results while the latter shows large positive realized P/L.From this, it can be seen that GBP/USD AI is gradually learning the market.

EUR/USD

Finally, let’s look at EUR/USD data.

Compared to USD/JPY AI, EUR/USD AI generally holds unrealized P/L larger, and compared to GBP/USD AI it tends to keep smaller realized losses while seeking profits.Additionally, like GBP/USD AI, it posts positive results in both market declines and rises, indicating it tends to hold positions forecasting which way the market will move next.

Notably, during March 2021 when the market fell sharply and then rose again, unrealized P/L remained largely positive and there were days with positive realized P/L as well.
In the two above currency pairs, P/L fluctuated with market moves, but EUR/USD AI maintained positive P/L even when the move reversed, which is notable.

On the other hand, although unrealized P/L is often positive, realized P/L did not always end positively, suggesting some AIs may be missing optimal exit timings.
If your AI agent often holds positions because it misses exit timings, helping it by evaluating trades may improve it.

In conclusion

This time, by comparing each currency pair’s P/L and rate, we examined what kinds of markets the AI agents tend to excel in.

Overall, it appears that many AIs trade with some awareness of trends, and as long as unrealized losses do not become negative, they tend to continue holding positions to seize opportunities.
Therefore, in low-volatility, range-bound markets, AI is less likely to place many new orders/close many positions, making it harder to accumulate profits.

Also, even in loss periods, there are areas where losses seem to diminish when similar situations recur. This indicates it is very interesting to observe AI agents learning and adapting to the market.

However, it is also true that there are situations where AI agents’ self-learning alone cannot handle everything.

The present data represent averages across many AI agents and do not necessarily apply to every AI agent.

For your own AI agent,it is advisable to carefully decide whether to educate it or to let its self-learning continue without intervention, by closely checking trading results.