I encountered a situation trading today which made me realize just how un-optimally I am trading and what difficulty is involved mentally calculating the multi conditional probabilities involved in trading.

Here's the scenario.  I'm long index futures.  We get a big run up midday and then we hover.  Now this particular strategy forces a hold for longer than this, but I am watching it with my discretionary eye saying, maybe I should take some profits off the table here.  What's worse is that we've come off the highs a bit and seem poised to go lower.  Below is the chart at the point of decision:

At this point I am thinking, well perhaps what I can do is exit now, and if price actually goes back up and breaks the former high I will get back in.  I will miss out on the profit between where I am now and where I would reenter, but it might be worth it if exiting now allows me to avoid any further downside erosion.  Also when I reenter I will put a stop in at where I just exited, a decent approximation of the bottom of this range.  Thus I could be introducing another loss if I reenter and the market turns around again.  Additionally, if I repeat the process more than once I could find myself incurring multiple losses this way.

Below is a chart detailing some of the outcomes.

Obviously there is no real way of solving this without my expected probabilities for each event taking place.  I think its somewhat complex when you have to put a percentage on
"run up to high, retrace and get stopped out, run up to high, run to anticipated profit".  It seems these things can be approximated by intuition.  However that's usually sub optimal.

I guess I'm writing to see how others would handle this, to spur discussion about sub optimal decision making, and perhaps even some discretionary vs systematic talk.  Hopefully this post does the job.  I'm all ears.

Hi opmtrader- reading your old posts you strike me as a fairly systematic trading guy. So my 2 cents would be to just follow your systems' established entry & exit rules and go with the historical odds. I think FDAXhunter mentioned somewhere that all systems go through "hot" and "cold" periods, so I guess the general advice would be to accept that

1 - all systems will do evidently "stupid" things at times (but then again you only know after the trade is complete)

2 - the way to correct it is by having other systems/models balancing your total portfolio

This sometimes results in one model exiting a long at 4245 just to have another model entering five minutes later at 4249. Ýou pay commission, spread etc. and it's frustrating for would they have somehow coincided in time, it would have generated an internal trade at midprice (no spread/no commission...).but I still wouldn't second-guess the decisions...

My personal (somewhat limited) experience with takeprofits/stoplosses/reentry etc. is that they don't add anything but whipsaws and ultimately lower the avg. profit and even worsens all stats that I intuitively felt I was enhancing. I remember though reading in market wizards that eckhardt or dennis reentered a position when it hit a new extreme (high or low) after being stopped out. So I guess it's a question of "taste"?

I am sure some of the more mathematically inclined np members could provide more formal arguments for/against the use of stoplosses/reentry strategies. Mine is soleley based on data/parameter-mining

cheers,

Nonius, amazing, I was thinking exactly this same thought only last week. Set up the data. I'd be interested to try it out. Incidentally, to pin my colours to the mast, I think we need to be incredibly modest in our expectations of what we can infer from data.

I think I should be further out than 5:2.

I actually did a mini version of this exercise last week and found that, even for quite simple process models, I couldn't reliably infer the model from the simulated data. This conclusion holds good and bad news for systems traders: bad news because, wow, if you can't recover a simple model on a spreadsheet then how on Earth can you recover *God's* invisible market model? Good news because of the implication for the weakness of the EMH.

Sorry dudes just a question: given one or more time series (like the one Nonius just posted) what sort of answer you guys should provide? Only if it's random or not?

hmm, b) is a useful criterion in a game like this where the process generating model is fixed. But it's not obviously useful in real trading, where trades that backtest well are often trades *discovered* by everyone and which therefore turn out unprofitable. b) therefore has the effect of putting a substantial wedge between this game and trading reality.