What's the minimum you can reasonably hold a position based on your data granularity?

[Rocko Bonaparte]

I'm trying to get an impression how long people here tend to hold their positions at the shortest, given the granularity of their data feed. I suck myself into ideas involving really short-term data, only to figure out that I'm trying to do too much with too far bars' worth of data.

Assuming I am working with 1-minute bars. At a minimum, I expect I would need:

1. one bar to figure out something is going on
2. another bar to open a position
3. another bar to realize I need to close the position
4. another bar to close the position

So I suppose--without limits or stops, at 1-minute bars I can't expect to do anything with less than four bars. But practically speaking, I should be thinking some multiple more. Like, at least 24 bars. And the whole reason I got myself some 1-minute data was to have the granularity to work in this range. And there I go, doing stuff in 4-10 bars.

Would any of you be willing to share how short-term you tend to get with your stuff?

 

[RangeTrader]

Some people go down to fifteen second bars... I find it unnecessary for the ES... 1m does just fine. Honestly the bars are not that important... Just the cycle pattern and the price levels are all that matters.

Sometimes trades last 1 minute... Sometimes five minutes for short term ES trades... Just depends upon how fast the market is moving and how long it takes to test the proper price levels...

 

[dom993]

The timeframe you use need to have enough resolution for you to "see" all the details of price-action required to implement the strategy.

Of course, that means than a given "pattern" will take several bars to form, and hopefully your trade will also take several bars & even more than that - the shortest trades should statistically be your losers, not your winners.

I use 100-volume on CL ... patterns usually form in in a couple hundred bars, and trades duration (in bars) from 5 to ~1500 for my DF20 system.


It seems you want to have a high frequency of trading ... just remember that on each trade, you pay commissions, spread/slippage, and those cost are better be small when compared to you average net / trade ... usually that means "longer trades".

 

[sheepsucker]

Quote from dom993:



Of course, that means than a given "pattern" will take several bars to form, and hopefully your trade will also take several bars & even more than that - the shortest trades should statistically be your losers, not your winners.

We seem to have very different strategies, my losers are almost 2 times longer than the winners.

 

[Rocko Bonaparte]

I assumed by shorter trades being loser was that they triggered stops or otherwise early exits.

 

[dom993]

Quote from sheepsucker:

We seem to have very different strategies, my losers are almost 2 times longer than the winners.

That system is a reversal system (picture this as taking the 1st pullback after a trendline break), trading one reasonably near target (avg Reward:Risk = 0.90) & one runner (avg Reward:Risk = 2.30) ... considering target-1 & runner as different trades, the stats are as follow:

For target-1 alone: (win% ~= 70%)
Wins : avg 62 bars / stdev 62 bars
Loss : avg 61 bars / stdev 54 bars

For runner alone: (win% ~= 50%)
Wins : avg 350 bars / stdev 97 bars
Loss : avg 249 bars / stdev 86 bars

The much longer average length in trade for runners/losers comes from the absence of any "break-even stop" mechanism ... stop is always in a logical place (for a long, under last HL), often times the stop remains below the entry (for a long) for quite some time after the target-1 is filled.

I would venture to guess that if you have shorter avg time in trade for your wins than for your losses, your system shoots for high win% using small targets (avg Reward:Risk << 1). Correct?

 

[sheepsucker]

Quote from dom993:

That system is a reversal system (picture this as taking the 1st pullback after a trendline break), trading one reasonably near target (avg Reward:Risk = 0.90) & one runner (avg Reward:Risk = 2.30) ... considering target-1 & runner as different trades, the stats are as follow:

For target-1 alone: (win% ~= 70%)
Wins : avg 62 bars / stdev 62 bars
Loss : avg 61 bars / stdev 54 bars

For runner alone: (win% ~= 50%)
Wins : avg 350 bars / stdev 97 bars
Loss : avg 249 bars / stdev 86 bars

The much longer average length in trade for runners/losers comes from the absence of any "break-even stop" mechanism ... stop is always in a logical place (for a long, under last HL), often times the stop remains below the entry (for a long) for quite some time after the target-1 is filled.

I would venture to guess that if you have shorter avg time in trade for your wins than for your losses, your system shoots for high win% using small targets (avg Reward:Risk << 1). Correct?

Generally yes, more correct on some of the points than others. Direction of the guess was of course right.
Eg. one system:
win% 75
RR: 0.85

One would think it would be easy to add a filter to tighten target after certain time to increase profits, but just keeping it simple as it is has worked best

 

[gmst]

Quote from dom993:

That system is a reversal system (picture this as taking the 1st pullback after a trendline break), trading one reasonably near target (avg Reward:Risk = 0.90) & one runner (avg Reward:Risk = 2.30) ... considering target-1 & runner as different trades, the stats are as follow:

For target-1 alone: (win% ~= 70%)
Wins : avg 62 bars / stdev 62 bars
Loss : avg 61 bars / stdev 54 bars

For runner alone: (win% ~= 50%)
Wins : avg 350 bars / stdev 97 bars
Loss : avg 249 bars / stdev 86 bars

The much longer average length in trade for runners/losers comes from the absence of any "break-even stop" mechanism ... stop is always in a logical place (for a long, under last HL), often times the stop remains below the entry (for a long) for quite some time after the target-1 is filled.

I would venture to guess that if you have shorter avg time in trade for your wins than for your losses, your system shoots for high win% using small targets (avg Reward:Risk << 1). Correct?

Its a rarity to see such mathematical approach to trading on ET. Kudos!

Sending you a PM.

 

[Rocko Bonaparte]

I just started posting again myself after a few years and I'm pleased to see what looks like a pleasant culture change. I suppose I'm partial to hanging out at the nerd table generally. I'm very glad to see those numbers on time frames. I threw out what I was working on since there really wasn't a clear pattern to me between the winners and losers.

I do have to ask though: what is up with those standard deviations? They seem huge!

 

[dom993]

On the general use of stats ... yes, I believe some stats are useful and do actually show something.

One of my favorite is to look at the distribution of MAE, separately for wins & losses. That tells a lot on the effectiveness of the entry placement.

For example, using the DF20 system / target-1 ... the MAE distribution for the wins shows without too much of a doubt that the entry placement has an edge ... ie. it has a visible & pretty consistent slant, the highest count of entries is for MAE=0.01 (1-tick) followed by MAE=0.02, the lowest count for MAE=0.27 (damn - that's no perfect, counts for MAE=0.28 & 0.29 are higher )


I encourage everyone to do this very analysis on their systems ... this is pretty much step-1 towards a better entry placement (when at all possible).