after a number of trades from the same strategy, why my winners win? why my losers lose?

tomdwan · May 1, 2018

Dear All,

Goodday.

Let say i have a strategy and i made 1000 trades from this strategy. Assuming i have 600 winners and 400 losers.

I wonder why do my winners win and why my losers lose? Its from the same strategy/setup. Can we blame my losers on randomness? If that's the case, my winners could be due to randomness too isn't?

Thank you.

Hotcakes · May 1, 2018

Markets trend. They just do. Try and develop strategies that capture market trendiness....

southall · May 1, 2018

tomdwan said:
Dear All,

Goodday.

Let say i have a strategy and i made 1000 trades from this strategy. Assuming i have 600 winners and 400 losers.

Thank you.

More like you will have 600 losers and 400 winners.
And even most of those 400 winners wont be very big.
The big easy to capture moves happen a minority of time.
Choppy moves happen most of the time.

tommcginnis · May 1, 2018

1) You are placing deterministic positions on a non-deterministic, probability-driven, "stochastic" process.

2) The success/failure of those positions will be subject to the same non-deterministic, probability-driven, "stochastic" process's distribution/behavior.

[Not asked for, but probably needed:]
3) It is comforting, but erroneous, to assume the stochastic nature of the environment is either constant, or unbiased, however both of these assumptions are regularly violated.

qxr1011 · May 1, 2018

tomdwan said:
Dear All,

Goodday.

Let say i have a strategy and i made 1000 trades from this strategy. Assuming i have 600 winners and 400 losers.

I wonder why do my winners win and why my losers lose? Its from the same strategy/setup. Can we blame my losers on randomness? If that's the case, my winners could be due to randomness too isn't?

Thank you.

make another 1000 trades and you will know the answer

jinxu · May 1, 2018

qxr1011 said:
make another 1000 trades and you will know the answer

A thousand trades is enough of a sample size to form a good conclusion. Plus he mention that this is a theoretical scenario. My opinion? The 600/400 is close enough to 50% that it is consider randomness. More homework needed. Hehe.

tommcginnis · May 1, 2018

jinxu said:
A thousand trades is enough of a sample size to form a good conclusion. Plus he mention that this is a theoretical scenario. My opinion? The 600/400 is close enough to 50% that it is consider randomness. More homework needed. Hehe.

I know: it's a busy afternoon and all but, he specified that the numbers were arbitrary chosen, and (we all) infer that n=1000 is sufficient to get a lot done. But let me oh so gently point out that n|fail = 400 is 150% of n|success = 600, and that ain't no random in a reasonable universe.

Xela · May 1, 2018

jinxu said:
The 600/400 is close enough to 50% that it is consider randomness.

You're joking (I hope)?! Over 1,000 outcomes, it's more than 6 standard deviations from the mean: they don't often come much more conclusively non-random than that ...

jinxu · May 1, 2018

I'm kind of thinking people are not using statistics in the correct way here...

The 600 and 400 win lose can easily swing the other way. Play roulette and sometimes you can get black 10 times in a row. Hehe

Kevin Schmit · May 1, 2018

jinxu said:
My opinion? The 600/400 is close enough to 50% that it is consider randomness.

If the trades are binary (equal risk and reward), then his result is significant at a better than 0.001 level. Chi^2 is around 20 and 95% confidence of true (population) expectation of > 55%. If this trading model is only one (maybe the best performing) of OP's many models, you'd need some type of multiple comparisons correction, but it's a strong result regardless.

OP could try building a separate model(s) to distinguish between likely winners an losers. Perhaps he could find indicators or features that discriminate between winners and losers. I would start with Fisher linear discriminant analysis on standard market features. I would also model longs and shorts separately (build two models). David Aronsen (author of "Evidence Based Technical Analysis") wrote and few papers and tutorials on this type of [what he called "filter"] model, you can probably find them by Googling.