In the previous article, we discussed that ACE Gold places importance on the number of samples when searching for past similar channels.Sample size is emphasized.

If there are only a few samples,

even if it shows “Failure rate 80%,”

it is dangerous to trust it just by that.

So, conversely,

Is more samples always better?

As development progressed, we began to question this here as well.

More data, more reassurance?

For example,

if 21 out of the past 30 cases recoiled.

The failure rate is 70%.

On the other hand,

if 700 out of the past 1,000 cases recoiled,

the failure rate is also 70%.

If it’s the same 70%, logically the 1,000 cases would seem more trustworthy.

I thought so too.

Therefore, even at ACE we have been developing to use as much past data as possible to perform statistics.

However, one problem arises here.

When was that 1,000 cases’ data from?

That is the issue.

Are the markets of 10 years ago and now the same?

The GOLD that ACE Gold targets has also seen market conditions change over a long period.

The magnitude of price moves differs.

The way trends appear differs.

Price levels themselves have changed significantly.

Thus, questions arise.

For example,

the past 100 cases in the last year

and

1,000 cases including many years ago

might be considered.

Statistically, the 1,000 cases offer richer samples.

But,

are 1,000 cases really superior as material to judge the current market?

I don’t think this is straightforward.

What ACE is doing now

Currently, ACE searches past price movements for channels with similar conditions and

uses what happened after that as statistics

to inform its analysis.

And now, the past data being used is basically weighted equally.

A similar channel that occurred a year ago is treated as one data point,

as is a similar channel that occurred much earlier.

If the conditions match, it is counted as one statistic item.

This is a very straightforward method.

However, as verification continued,

the question arose: “Is it really okay to treat old data and new data as the same single vote?”

This is not simple.

Throwing away old data isn’t the only option

Then,

you might think, “Just use recent data.”

But that isn’t easy either.

If you shorten the period, it becomes closer to the current market.

However,

the sample size decreases.

For rare channel shapes, there may not be enough cases.

Conversely, if you lengthen the period,

the sample size increases.

But then it includes market environments that are quite different from the present.

In other words,

prioritizing novelty reduces the sample size.

Prioritizing the sample size increases old data.

This presents a rather interesting challenge when building a trading system.