How to Find Winning Anomaly-type EAs! [From a Developer's Perspective]

This time, we will explain what a profitable anomaly-based EA is, its characteristics, and how to choose it.

■ What is an anomaly-based EA?

An anomaly-based EA refers to an automated trading program that detects regularities in the forex market based on historical exchange rates and trades by taking advantage of those regularities.

In short, an anomaly-based EA is an automated forex trading program that uses the regularities of the currency market.

Specifically, when there is a regularity at certain times or dates in the market, it uses that anomaly to buy and sell.

Below are the features and representative examples of anomaly-based EAs.

■ Tokyo Fixing Trade EA

Tokyo fixing trading is a method that takes advantage of temporary large moves in the market during the “fixing” time used by financial institutions as the reference rate.

The fixing time is considered to be 9:55 am Japan time, and trading with awareness of this time is said to yield efficient wins.

The logic of the fixing-trade EA varies by developer.

Trade frequency varies by EA, from those that trade daily to those that enter only a few times a month.

Generally, EAs that limit entry to particular days (GOTO days) such as month-ends or weekends tend to show better performance.

■EA targeting the London Fix

London fixing refers to the time when the London market determines the fixing rate.

In London time it is 16:00, which is 24:00 Japan time in daylight saving, or 25:00 in winter.

There is a well-known anomaly that the pound tends to strengthen around month-end London fixings.

The basic strategy is to target the rise before the London fixing and attack the period when the pound buying pressure increases.

Example: in summer time, go long on the pound-dollar or pound-yen at 23:00 Japan time; in winter, at 24:00 Japan time, and close after one hour.

Example: buy the pound at month-end.

Based on historical statistics, the pound tends to rise at the month-end London fixing, so buying the pound then can aim for profit.

Reasons cited for these statistics include:

Month-end settlement: many companies settle accounts at the end of the month, increasing demand to buy pounds.

Window dressing: some financial institutions buy pounds to present assets more attractively toward month-end settlements.

Psychological factors: investors recognize the tendency for the pound to rise at the month-end London fixing and may follow by buying the pound.

During London fixing, pound-buying and during Tokyo fixing, dollar-buying are said to occur.

There are other reasons that are not clearly understood but are included as anomaly-based EAs.

■ Comparison with discretionary trading

With discretionary trading, you can reflect current market conditions rather than relying solely on past data.

EAs cannot trade in a way that suits the current market unless they are updated or their parameter values are changed.

In other words, if the logic no longer fits the current market, the EA will not win.

■ Characteristics of a winning anomaly-based EA

A winning anomaly-based EA analyzes massive past data to extract reliable patterns.

The more accurate this analysis, the higher the success rate of trades.

I developed an EA called “London Time Trade USDJPY.”

London_Time_Trade_USDJPY product page

It enters new positions on the day’s counter-trend just after 16:00 Japan time and closes within the day.

Since its release on May 24, 2021, it has experienced ups and downs, and as of May 2024 it is marginally profitable.

■ Why anomaly-based strategies may not work

London fixing is said to have less influence on exchange rates in recent years.

Also, there are several possible reasons why Tokyo fixing trading has been underperforming since 2023.

Reason 1: Decreased trade volumes and the development of financial markets may reduce the demand to buy dollars based on Tokyo fixing.

In the past, exporters would fund dollars around Tokyo fixing for month-end trade settlements, but such transactions are thought to have decreased recently.

Reason 2: Tokyo fixing trading has become known to more traders than before.

As a result, many orders cluster during Tokyo fixing time, potentially diluting the anomaly effect.

Reason 3: Globalization and electrification of financial markets have increased Forex rate volatility in shorter timeframes.

Thus, anomalies effective only at specific times, like Tokyo fixing, may have weakened.

There is a need for countermeasures against the biggest weakness of anomaly-based EAs, where the logic stops working.

■ Measures to win stably

Measure 1: Risk management

No matter how superior an anomaly is, not all trades will be successful.

Winning EAs include risk management features to minimize losses.

Appropriate stop-loss and take-profit settings are essential.

Measure 2: Forward results verification

Backtesting on past data is important to evaluate anomaly-based EA performance.

However, backtesting alone cannot fully predict future market conditions.

Forward results showing a consistent positive performance are the most important factor when choosing an EA.

Measure 3: Analyze the basis of the anomaly

A baseless anomaly is unlikely to have future potential.

Since the basis for anomalies comes from fundamental analysis, it provides valuable information for judging whether it remains effective in the current market.

Measure 4: Is the EA’s logic transparent?

Black-box EAs carry higher risk, so choose ones that clearly explain how profits are generated.

If the logic is not clear, the EA may be optimized for past markets only.

Publicly disclosed logic allows users to understand entry and exit timings, aiding the decision to run the EA.

【Conclusion】

・Diversify your portfolio with various anomaly-based EAs.

・If the anomaly stops working, losses begin. Remove EAs with poor forward results from your portfolio.

・If the logic is disclosed, consider whether to run it including fundamental analysis.

■ Advice from an EA developer

I have developed dozens of anomaly-based EAs, and they generally perform worse than EAs based on technical indicators.

The reason is that their logic is designed to fit past market conditions and lacks an edge for future markets.

Why does this happen? EAs that do not provide clear justification for their logic are rarely profitable in forward tests.

Also, anomaly-based EAs that fail in forward results should not be used.

I hope this is helpful for your investing.