If you tried to examine everything, the calculation did not finish

Even AI tools cannot compute infinitely. As the amount of data we handle increases, so does the time required. It should be obvious, yet it was something I had overlooked until I actually encountered it.

Previous time → 【Episode 1】I finished examining the time axis in between, so this time I tried something even shorter.

A Recap of Up to the Previous Time

Last time, we announced that we would tackle shorter time axes that had been left untouched so far (M15, M30). We conveyed that while we had a hunch that shorter time axes would make validation stricter, we decided to start this validation anyway.

When I Tried to Move as Usual, It Didn’t Start

In our previous validations, we directed the in-house tool “IKLab AI Strategy Builder” to use data from the entire period, and simply ran it as is. There were no issues whether it was the 5-minute chart (M5), the hourly chart (H1), or the 4-hour chart (H4).

However, when I tried doing the same with even shorter time axes (M15, M30), I hit an unexpected wall.I realized that the computation per pair took longer by an order of magnitude compared to before.

【Image: 33_data_vs_time_tradeoff.png to be inserted here (owner will paste manually)】

Why Didn’t the Same Method Work as Before

To sort out the reasons, it turned out to be surprisingly simple. As the time axis becomes shorter, the number of price-movement data points within the same period increases. Our tool performs a robustness test that stubbornly checks whether each candidate is real from five different angles every time a candidate is found. Because this verification process involves re-running backtests for each candidate, the more data there is, the longer the verification takes in proportion.

In other words, shortening the time axis means “viewing the same period in finer detail” and simultaneously “increasing the amount of computation required for validation.” Things that were irrelevant with longer time axes became a non-negotiable obstacle at shorter time axes.

Giving Up on “See Everything” for Now

When I actually tested, progressing the search with data for the entire period alone took a vastly longer amount of time just to examine a single currency pair. Moreover, we would need to perform this for 29 currency pairs. If we continued with this method, it would be unrealistic to finish in a practical timeframe.

So we faced a choice.Do we insist on “seeing everything,” or switch to “see what can be checked now”?After weighing, we chose the latter. Instead of examining the entire period as it is, we decided to narrow the validation to a recent, fixed period.

【Image: 33_full_vs_feasible.png to be inserted here (owner will paste manually)】

Narrowing Down and Getting It to Work to Some Extent

Specifically, to keep the calculation within roughly the same amount as the previously issue-free time axis (4-hour chart), we adjusted the number of data points we target. The time axis itself is shortened, but the amount of data handled is kept at the same scale as before.

As a result, we were able to complete calculations within a realistic timeframe as before. However, this comes with a significant trade-off.While the time resolution (how finely we observe price movements) has increased, the range of validation on the calendar that we can go back to is shorter than before—in other words, we can no longer say we have seen the entire history.

【Image: 33_narrow_down_flow.png to be inserted here (owner will paste manually)】

I’ll Honestly Carry This Choice Forward

This “narrowing” is a technical measure that has not appeared in any previous installments of this series; it is unique to this time. Therefore, I intend to be transparent about this fact in future posts as well. It would run counter to the ethos of this research institute to show only the result “short time axis was examined” and reveal later that “in reality, only part of the calendar was viewed.”

Details such as exactly how short the period was, and which currency pairs were involved, will be shared in upcoming posts.

Disclaimer

• The contents of this note do not constitute recommendations to buy or sell specifically, nor investment advice. It is for information sharing only.
• Backtesting results based on past data do not guarantee the same results in the future. Markets are always changing.
• Whether you actually trade or not is your own responsibility.

Next time, I’ll share experiences where even though the narrowing produced seemingly good numbers, not a single one passed. If you’d like, please follow so you can receive the next article as well.

※Currently, the official LINE account is in preparation. We will announce here as soon as it is established.

Next: 【Episode 3】There were many promising numbers, but not a single one remained