How far into the future can we really predict?
Therefore, if someone aims to capture large moves using only one or two positions, the forecast must often extend over many hours, days, or even weeks
Think about it this way...
If you use a weekly chart, several weeks are merely the next several bars.
You might want to do a search in the CodeBase for the Nearest Neightbor Weighted Coefficient indicator.
As a retail trader you are out the loop and can't see behind the curtain unless you are one of The Donalds pals you don't get a heads up . Too many variables and goings on you will never be able to go on a prediction run and if you do it is riding luck .There is a fine line from , I knew that would happen to What just happened .
What time horizon are we talking about here?
As we know, "many" and "few" are relative terms. You can have many insignificant variables or a few extremely important ones. That becomes a different discussion altogether.
But if you mean that it is impossible to reliably see far ahead (days or weeks), then yes — I agree, and that is exactly what I was writing about. In my view, expecting deterministic long-range prediction is a utopia.
And my next question to anyone who claims it is possible is always: show me the results. If they differ significantly from a random 50/50 outcome, then perhaps you have discovered some genuinely new property of reality.
So far, no one has been able to demonstrate that to me.
That is why I believe that looking far into the future with certainty is impossible. The future emerges from the present, and absolute determinism is not achievable.
So yes — I agree with your point, and that is exactly what I have been arguing.
Think about it this way...
If you use a weekly chart, several weeks are merely the next several bars.
You might want to do a search in the CodeBase for the Nearest Neightbor Weighted Coefficient indicator.
Perhaps you misunderstood my question.
If you want to discuss how to think about it, I can happily expand on my own view using an n-dimensional vector-space description. That part is actually quite clear to me.
My question is much simpler.
I'm not a beginner. What interests me is the opinion of people who believe it is possible to forecast the market.
When you open a trade at time (t_0), based on whatever assumptions you have at that moment, after how much time do you personally conclude that the information you had is no longer valid because the system itself has changed?
One hour?
Two hours?
A week?
That's really the whole question.
As for the mathematical picture, I could even share a video of how I visualize the market as an n-dimensional hyperspace. Of course, we cannot physically visualize more than three dimensions, but even that already gives a much better intuition that reality is far more complex than our usual charts suggest 🙂
What time horizon are we talking about here?
As we know, "many" and "few" are relative terms. You can have many insignificant variables or a few extremely important ones. That becomes a different discussion altogether.
But if you mean that it is impossible to reliably see far ahead (days or weeks), then yes — I agree, and that is exactly what I was writing about. In my view, expecting deterministic long-range prediction is a utopia.
And my next question to anyone who claims it is possible is always: show me the results. If they differ significantly from a random 50/50 outcome, then perhaps you have discovered some genuinely new property of reality.
So far, no one has been able to demonstrate that to me.
That is why I believe that looking far into the future with certainty is impossible. The future emerges from the present, and absolute determinism is not achievable.
So yes — I agree with your point, and that is exactly what I have been arguing.
[A]fter how much time do you personally conclude that the information you had is no longer valid because the system itself has changed?
I'm afraid that my answer to your refined question (thank you for that), won't be very satisfactory to you...
My preferred chart structure is a custom chart and specifically, a Renko chart. As you probably already know, Renko charts are "timeless." I've enclosed that word in quotes because Renko bricks still have timestamps, even though the incremental scale of the time (x) axis of the chart is heavily distorted based on the speed in which price moves. If the bricks form quickly, the timestamp increments of the bricks are condensed. If the bricks form slowly, the timestamp increments of the bricks are expanded.
Therefore, time is largely irrelevant to my EA's─at least in terms of the validity that you describe. Of course, I generally implement an entry time filter that goes on and off 4 times per day, a daily filter, and/or a monthly filter. I also shut down entries an hour prior to weekly market close, and forcibly close all positions a half hour prior to weekly market close. (This is a single position trading strategy).
I'm not a big fan of adaptive/machine learning code. I backtest over many years, and analyze each year in the backtest report for any potential black swans. If I model a Renko chart based on the M5 timeframe, and test on it for 5 1/2 years, the statistics generated don't lie─assuming that the user can properly and accurately backtest.
Having said that, I did get some good anecdotal signals from the NNWC indicator for a bit.
Well yes I was talking days and weeks but all it takes is a tweet from a certain someone and that would be minutes as a knee jerk reaction .
Let me clarify my position once more.
I personally see no reason to believe that reliable long-term prediction (days, many days, weeks, let alone months) is possible. If you asked me directly, I would say there is no such possibility. In that sense, your position is actually very close to mine.
There is also a mathematical aspect.
The quality of any forecasting method is ultimately measured by its win rate. That metric tells us whether the original prediction contained genuine information or whether the result was simply consistent with chance.
Take a coin toss as the simplest example. The larger the sample, the closer the observed frequency approaches 50/50. In my opinion, something similar happens with many trend-following approaches. Essentially, they observe price movement, manage exits well, and occasionally let winners run. But over hundreds or thousands of trades, the directional prediction itself tends to stay close to random.
However, you mentioned something interesting: skill.
What exactly is skill?
This is where I think AI becomes interesting. In principle, it can replace part of what we usually call human "trading skill."
One of my own AI models has now executed more than 11,000 trades while maintaining a win rate above 95%.
That is precisely why I am interested in this discussion. My intuition is that only the near future can be "seen" from the present state. Beyond that, reality continues to evolve, and new information appears that simply did not exist at time (t_0).
There is another thought that I find especially fascinating.
Even the past cannot be perfectly reconstructed. No matter how much additional information you collect later, there are always states of the system that can no longer be uniquely recovered.
To me, that is one of the most interesting properties of reality itself.
I'm afraid that my answer to your refined question (thank you for that), won't be very satisfactory to you...
My preferred chart structure is a custom chart and specifically, a Renko chart. As you probably already know, Renko charts are "timeless." I've enclosed that word in quotes because Renko bricks still have timestamps, even though the incremental scale of the time (x) axis of the chart is heavily distorted based on the speed in which price moves. If the bricks form quickly, the timestamp increments of the bricks are condensed. If the bricks form slowly, the timestamp increments of the bricks are expanded.
Therefore, time is largely irrelevant to my EA's─at least in terms of the validity that you describe. Of course, I generally implement an entry time filter that goes on and off 4 times per day, a daily filter, and/or a monthly filter. I also shut down entries an hour prior to weekly market close, and forcibly close all positions a half hour prior to weekly market close. (This is a single position trading strategy).
I'm not a big fan of adaptive/machine learning code. I backtest over many years, and analyze each year in the backtest report for any potential black swans. If I model a Renko chart based on the M5 timeframe, and test on it for 5 1/2 years, the statistics generated don't lie─assuming that the user can properly and accurately backtest.
Having said that, I did get some good anecdotal signals from the NNWC indicator for a bit.
The topic of time itself is actually a huge one.
In fact, I would argue that time is not a fundamental quantity by itself. We can go very deep into physics here, but I think everyone would agree that what we really observe are events. We simply divide those events into intervals because things change, appear, disappear, and evolve.
A car moves. To describe that movement we introduce time. Then we take the first derivative and call it velocity. Then the second derivative and call it acceleration. In other words, we are performing mathematical transformations to describe processes.
Time itself is a fascinating philosophical question. Without our agreed "tick-tick" seconds, we simply wouldn't have a practical way to describe change mathematically.
So I think the question of what time really is remains open.
What interests me even more, though, is your point about machine learning.
Here I completely disagree.
I am strongly in favor of continuous learning. Personally, I think static pre-training alone is a dead end, while dynamic adaptation is a much more promising direction.
You mentioned that your system has survived for 5–6 years.
May I ask:
-
What is its win rate?
-
Profit factor?
-
How many instruments does it trade?
-
How frequently does it trade?
-
On what time horizons?
Without those numbers it is difficult to judge the system objectively.
And if those metrics are inconsistent—good in some periods and poor in others—then perhaps the system still has room for improvement. 😉
Let me clarify my position once more.
I personally see no reason to believe that reliable long-term prediction (days, many days, weeks, let alone months) is possible. If you asked me directly, I would say there is no such possibility. In that sense, your position is actually very close to mine.
There is also a mathematical aspect.
The quality of any forecasting method is ultimately measured by its win rate. That metric tells us whether the original prediction contained genuine information or whether the result was simply consistent with chance.
Take a coin toss as the simplest example. The larger the sample, the closer the observed frequency approaches 50/50. In my opinion, something similar happens with many trend-following approaches. Essentially, they observe price movement, manage exits well, and occasionally let winners run. But over hundreds or thousands of trades, the directional prediction itself tends to stay close to random.
However, you mentioned something interesting: skill.
What exactly is skill?
This is where I think AI becomes interesting. In principle, it can replace part of what we usually call human "trading skill."
One of my own AI models has now executed more than 11,000 trades while maintaining a win rate above 95%.
That is precisely why I am interested in this discussion. My intuition is that only the near future can be "seen" from the present state. Beyond that, reality continues to evolve, and new information appears that simply did not exist at time (t_0).
There is another thought that I find especially fascinating.
Even the past cannot be perfectly reconstructed. No matter how much additional information you collect later, there are always states of the system that can no longer be uniquely recovered.
To me, that is one of the most interesting properties of reality itself.
The past in trading is no marker , it is a joke .I don't even bother with back tests . How ridiculous to say this happened in the past in a fast moving unpredictable environment. Retail trading in its entirety is fascinating, it is the greatest invention since roulette to play on every human emotion and rob you fast , if you are greedy or slow if you are more adept !! .
I wasn't talking about backtesting.
I mean something much broader.
Not taking old data and applying it to new data. And not simply projecting the past into the future.
No, something different.
Humans naturally think with time moving forward. But try to follow this thought.
Imagine running the film backwards.
At first glance, it seems that predicting the past should become easier.
Surprisingly, it doesn't.
Even when moving backward, we can keep adding more and more information—information that we now know but wasn't available before. Yet there are many studies showing that this still does not allow us to reconstruct the distant past perfectly.
So my point is broader than trading.
Long-range prediction appears to be fundamentally limited in both directions: from the present into the future, and even from the present back into the past.
As for the market "robbing" people...
I think that's a bit too pessimistic. 🙂
Yes, an overwhelming majority of retail traders lose money. Depending on the study, it's often more than 90%.
But "more than 90%" is still not "everyone." There are exceptions, and those exceptions are exactly what make the subject so interesting.
- Free trading apps
- Over 8,000 signals for copying
- Economic news for exploring financial markets
You agree to website policy and terms of use
Hello everyone,
I would like to ask what I believe is a very interesting question about the predictability of the future in financial markets.
As far as I know, there is still no definitive scientific consensus on whether the market is fundamentally a probabilistic system—where probability-based forecasting is the most appropriate approach, much like weather forecasting—or whether there exist relatively stable market structures that persist long enough to allow reliable prediction of specific future price levels.
Without going deeply into the mathematics, I would like to ask a practical question:
How far ahead has anyone here been able to forecast market behavior with consistent confidence?
One important clarification.
A pip is simply a unit of price movement. However, the number of pips that can realistically be expected is naturally limited by how prices actually move. In other words, expecting a market to move hundreds of pips within a few minutes is, in most cases, an extreme event rather than the norm.
Therefore, if someone aims to capture large moves using only one or two positions, the forecast must often extend over many hours, days, or even weeks.
So my main question is:
Does this actually work?
Has anyone here been able to make consistently successful forecasts several days ahead from a given moment (t_0)?
Personally, I suspect that predicting the distant future with certainty is impossible simply because of the nature of reality itself.
In other words, the future is probably emergent rather than deterministically computable over long time horizons.
At the same time, this does not rule out the existence of local stable structures, where the probability distribution of future outcomes becomes significantly biased toward certain scenarios for a limited period of time.
I would be very interested to hear your thoughts.