Forum > Trading-related Topics

Trading psychology vs. statistical edge

(1/4) > >>

PipMeHappy:

Kaitsu:
Interesting vid!

I was pleased to see he is also using an hourly chart, which is my personal favourite.

If I understood correctly, he is gauging statistical edge by comparing the results of his own discretionary use of a strategy with an automated version using the same strategy.

I tend to do the same but it is not always so simple as it might sound!

The main difficulty in trying to achieve any meaningful statistical comparison is that I am not always able or willing to actually trade every signal given by the strategy. This is especially so with hourly charts. Sometimes a strategy trade signal occurs overnight or on a day when I have other things to do, or prior to a significant event that I don't wish to trade through. One could choose to only compare those trades actually taken and ignore the other automated system trades but that then tends to distort the theoretical performance of the strategy as a whole.

Also, there is a tendency to start trying to outsmart one's own system which can lead to some bad habits such as anticipating a signal before it happens - only to find that the signal doesn't actually occur! I have that tendency! My strategy is based on hourly closes and there is always a temptation to pre-empt the close and try to enter at a better level during the candle - only to see the candle reverse direction before the end of the hour! :D 

On the other hand, there are legitimate ways to fine tune the results and produce a personal edge relative to the raw strategy. For example, statistically the following candle may usually retrace 10-30 pips from the previous candle close. This means one can often get a better entry level which can be then used in  a variety of ways such as a more distant stop or nearer target level for same R:R as the automated version, or a bigger profit, etc.

Another problem area is the size of trades. I tend to vary my position size depending on a subjective chart "feel" such as general volatity levels and may also close part or all of these positions entirely depending on how things feel e.g. if it seems that the market is struggling to reach those last few pips to my target then I will close it.

When there are a number of such variations, especially with variable position size, there is no meaningful statistical comparison with the automated results apart from the total pips earned at the end of a period such as a month-end.

There again, maybe I am looking at this the wrong way!  ;D 8) 

PipMeHappy:
@Kaitsu indeed a backtest should not be compared to the discretionary results as much as being used as a check on the validity of a discretionary idea. In short: one could make the discretionary strategy signal test one made purely of a series of binary rules, including elements of volatility and position size, making it as close to what we do as point-and-click discretionary traders.

I think this is very much where I want to head to next, i.e. looking at my discretionary trading to try defining the approach in terms that could be replicated in a signal test and that a computer would understand in binary terms.

I think Mathew Verdouw's work at Optuma has really opened my eyes to what I need to do next, which of course - as you may well know - is a lot of work!

https://www.optuma.com/videos/signal-testing-2/

Caesar:
Nice one there Francesco, lot of good stuff and clear presentation as well.
Agree with Kaitsu, I've tried outsmarting myself and burned my fingers so I'm not likely to tinker with something that currently works, even if I could theoretically improve.

I do like how you're constantly looking to improve your knowledge base, you're already well ahead of me from what I can see.

Enjoy your day guys,
Caesar

Kaitsu:

--- Quote from: PipMeHappy on October 18, 2020, 03:45:09 PM ---@Kaitsu indeed a backtest should not be compared to the discretionary results as much as being used as a check on the validity of a discretionary idea. In short: one could make the discretionary strategy signal test one made purely of a series of binary rules, including elements of volatility and position size, making it as close to what we do as point-and-click discretionary traders.

I think this is very much where I want to head to next, i.e. looking at my discretionary trading to try defining the approach in terms that could be replicated in a signal test and that a computer would understand in binary terms.

I think Mathew Verdouw's work at Optuma has really opened my eyes to what I need to do next, which of course - as you may well know - is a lot of work!


--- End quote ---
I watched the Optuma video and it certainly raises some interesting issues. I cannot claim to have fully appreciated the process going on there just from one viewing of this one video, but my first thought was that it only tells the performance of a particular strategy compared with a benchmark (in this case the SPX). You can adjust the parameters such as the stoploss but it does not appear to actually try to optimise the settings by running the test across a range of parameters from, say, X through to Y. So a strategy may well look OK but is it actually running at its best possible?

However, is this still not a form of backtesting? And does it not require that the strategy would have been run with precisely defined and totally unchanged parameters throughout the period of years included in the tests? That would be a tough order for most traders to meet!

I would find this very restrictive in my personal trading! But it would be understandable for a fund management needing to demonstrate to a client how a specific fund is performing. Personally, my "signals" are not that specifically defined. My strategy calls for a certain set-up at the end of a candle but that set-up can last for several candles before actually entering depending on a number of rather non-specific factors.

I have a problem with all these types of statistical analyses because the performance of any rigidly defined strategy is rigidly defined by the market itself and not the actual parameters of the strategy. And we all know how infinitely varied market movements can be especially when we factor in the increasing impact of algos and programmed trading. For this reason I have also always refrained from defining monthly targets and comparisons, because the same strategy can work very differently in different months and it is the market that determines this and not the strategy itself. Maybe over a longer time period these differences even themselves out but surely not always! For example, I can imagine a strategy designed for Crude Oil would have worked very differently in earlier years compared with the current and foreseeable market?

But some kind of benchmarking or comparative performance studies is certainly necessary where large sums are involved and/or when trading other people's money!

As you say, this is a big subject with many considerations and approaches and I guess for any trader the first task is to define what they actually need/want to know about their trading performance and their strategies and then to select what statistical studies help shed light on the subject for them.

But I am probably seriously understating the importance of quantitative studies, for which I sincerely apologise! :D   

Navigation

[0] Message Index

[#] Next page

Go to full version