TSR - resuscitating trading systems - page 5

New comment
 

While pulling the bar, I came up with an absolutely clear method of diversifying several strategies:

  1. We will consider as a separate TS even the one with different values of input parameters.
  2. Out of all the TS we left only those that are profitable.
  3. For each TP of the corresponding TS we found a linear regression - a straight line, which goes upwards (all are profitable).
  4. We subtracted its regression from each BP of profit and then divided the difference by the rightmost value of the regression (reduced to the same scale).
  5. Accordingly, the resulting GRs have zero MO (sample).
  6. We discarded those GRs (respectively, the TCs) that have a variance above a certain threshold (called risk), i.e. those that are too wide in the channel.
  7. Now with the remaining BPs we have done a reverse scaling - we have multiplied by the previously mentioned right regression value.
  8. Now we need to add up these BPs so that the variance becomes minimal. This will then give such an effect that lows in the profit graph of one TS will overlap with rises of another TS.
  9. Weighting coefficients all should be greater than zero and give one in total. (almost a classic task of Markowitz portfolio).

That is all, we have a certain set of TS with their weight coefficients. We respectively crossed all these TS and obtained the most diversified Super-TS with an absolutely precisely defined level of risk (see point 6).

 
hrenfx:

1. The proposed method has nothing to do with what I wrote in point 3.

2. Who cares if they are losing on the OOS, as long as they seem to be losing. That's what statistical arbitrage is all about.

1. The proposed method highlights positively correlated signals and suppresses negatively correlated signals. Isn't that a way of finding correlations? Then we are somewhere off on the terminology.

2. Maybe I do not understand the method of such trading. Can you elaborate? How to statistically arbitrage two sequences of trading signals for the same instrument?

 
voltair:

As for the EA, time will tell what it is worth.

The EA itself may not be worth anything, it is worth the idea it demonstrates and it is doing a great job with that. What else is required?
 
hrenfx:
That's it, we got a specific set of TS with their own weight coefficients. We respectively crossed all these TS and obtained the most diversified Super-TS with an absolutely precisely defined level of risk (see point 6).
You probably will not believe it, but I did it. The results are good, sometimes even impressive. But still even such Super-TS do not give guarantees. And it is necessary to apply other methods.
 
hrenfx:

While pulling the bar, I came up with an absolutely clear method of diversifying several strategies:

  1. We will consider as a separate TS even the one with different values of input parameters.
  2. Out of all the TS we left only those that are profitable.
  3. For each TP of the corresponding TS we found a linear regression - a straight line, which goes upwards (all are profitable).
  4. We subtracted its regression from each BP of profit and then divided the difference by the rightmost value of the regression (reduced to the same scale).
  5. Accordingly, the resulting GRs have zero MO (sample).
  6. We discarded those GRs (respectively, the TCs) that have a variance above a certain threshold (called risk), i.e. those that are too wide in the channel.
  7. Now with the remaining BPs we have done a reverse scaling - we have multiplied by the previously mentioned right regression value.
  8. Now we need to add up these BPs so that the variance becomes minimal. This will then give such an effect that lows in the profit diagrams of one TS will overlap with rises of the other TS.
  9. Weighting coefficients all should be greater than zero and give one in total. (almost a classic task of Markowitz portfolio).

That is all, we have a certain set of TS with their weight coefficients. Accordingly all these TS were crossed and obtained the maximum diversified Super-TS with exactly defined level of risk (see point 6).

This is exactly what Reshetov was dealing with in his previous project.

Here (in this topic) the idea is slightly different. Use logical multiplication instead of logical addition of signals. It is assumed (and practice confirms) that a considerable part of noise will be filtered out.

zy. Correction. Instead of arithmetic addition. // further on in the text.

 
MetaDriver:

1. The proposed method selects positively correlated signals and suppresses negatively correlated signals. Isn't that a way of finding correlations? Then we are somewhere off on the terminology.

I don't want to discuss the method proposed by the topicstarter any more.

2. Maybe I do not understand the method of such trading. Can you elaborate? How to statistically arbitrage two sequences of trading signals for the same instrument?

You can trade any number of sequences of trading signals for any number of instruments.

I am answering the question posed:

  1. Two BP Profits are correlated with each other.
  2. It means that their certain difference will hang in a horizontal channel.
  3. So you trade this channel like any flat.
 
voltair:
You probably won't believe it, but I've done it. The results are good, sometimes even impressive. But still, even such Super TCs do not provide guarantees. And you need to apply other methods.
What methods?
 
MetaDriver:
This is roughly what Reshetov did in his previous project.
A link?
 
hrenfx:

You can trade any number of sequences of trading signals on any number of instruments at once.

To answer the question posed:

  1. Two BP profits are correlated with each other.
  2. It means that their certain difference will hang in a horizontal channel.
  3. So you trade this channel like any other flat.
Unfortunately I don't know how to trade the profit/equity curve. Can you teach me? I honestly don't get it.
 
voltair:
You probably won't believe it, but I've done it. The results are good, sometimes even impressive. But still even such Super TS do not give guarantees. And you need to apply other methods.

The trick is that as the window moves (interval), even when our TC set does not change, the scales float. Well, the set of TCs also has to change.

This is the kind of self-adaptive Super TS that needs to be statistically investigated. I agree it is much better than all other proposed options, but from my point of view it is better not to deal with such diversification of TS (which in fact are multivariable functions of financial instruments, i.e. simply convert price BPs into Profit BPs), but go straight to searching of interrelations between initial data - price BPs. And trade exactly those relationships, which are definitely not contrived, but economically sound.

12345678910111213
New comment
Reason: