Frosty's trading bot goes live part 2

austinp:

I agree to most, but not all, of what you have written.

One simple question to understand where you're coming from...

When you were running / trying to run bots... how many systems and how many markets were you trading live, on any given time?

Another comment:

I took sometime reading a few posts Frosty wrote when he started in ET (Dec. 2005)...

It seems to me he hasn't learned much since then...
 
Quote from TSGannGalt:

You're thanking goodness that he lost money???
More...
No, I'm happy he didn't decide to focus on a contract that was doomed to failure (apparently).

Not sure what relevance that statement has to a thread devoted to automated trading ER2 though...
 
after following this thread, i must say ima bit discouraged. not because i have not made progress w/ my ats ..i have. it's just sounds as if i have no business attempting it. anyways, i assume that the only way to find that out is to step out there.. on the ledge, and see.
bb
 
Quote from frostengine:

Good news is I have entered the forward testing period with several new strategies..... so one of them will be going live in a few weeks.
More...

Are you changing anything about your methodology in response to the failure this time? I mean, I assume the bot you just terminated passed its forward testing phase.... Maybe the criteria you use to determine that a bot is fit to run need to change?
 
I have a few changes. #1 I am going to optimize for lower draw down instead of max PL.

Also, the next strategy to go through will NOT be a direct decendent of my previous strategies. I am going for a fresh strategy this time.
 
<i>"When you were running / trying to run bots... how many systems and how many markets were you trading live, on any given time?"</i>

The only profitable success I had in system bots was FX markets, GBPUSD specifically. Of all the markets = symbols I coded and wrote endless hours and computations for, that combo by far was the best.

I would strongly suggest anyone interested in system writing begin with FX in general and GBPUSD in particular. From there the real-time implementation has hurdles to overcome: round the clock trade execution, where to reside system code (home or broker desk) and factors like that.

As an example of what's possible in reality, here is a snapshot of one GBPUSD system that's been walking forward untouched since September 2004. Nearly three years of live data with zero tweaks or optimizations. Five years of data total, averages +$18,600 per single contract (no compounding) over that period of time.
 

Attachments

... now here's the rest of our story. How many traders are willing to live thru the drawdowns without changing a thing? Answer is, darned few. I ran this model thru a battery of tests via Rina and saw it was more robust than any others I'd ever tested.

Specifically, the software takes all recorded trades and scrambles their order at random. Then it trades forward in mock fashion and sees how a completely different sequence of market movement would have affected bottom line.

This model(family) tested out profitable to a high degree. It has recorded numerous months of drawdown, but holds up over time.

By the same token many other systems written flopped in frightening fashion when shaken down. That included systems written with moving averages and RSI. This model is written on volatility (open) breakout study, which is likewise the best tech analysis tool I found for FX system building.

Without a battery of shakedown tests per monte carlo simulation, the MA and RSI models looked much better than the volatility breakout in TS back-testing alone. But... those models designed on oscillators all augured straight into the ground when walked forward in real-time.

I'd guess that is much like what frosty has experienced. Back-tested results when building a system are a necessary foundation, but too much false info there for holding up in real time forward.

As for pure intraday models? They operate on a much thinner edge than this. Retail traders with low-level software need the highest level skills, lots of time tested (and lost) along with real dollars risked (and lost) to find answers.

Is it any wonder the big firms hire pure quants and equip them with million-dollar plus software to operate? Whom amongst us are capable of playing in their auto-bot sandbox?

Darn few
 

Attachments

Quote from frostengine:


Also, the next strategy to go through will NOT be a direct decendent of my previous strategies. I am going for a fresh strategy this time.
More...

frost,
There seems to be a number of people here who are willing to help you, if you would just let them.

A suggestion; since you're abandoning your last strategy, why don't you disclose it and maybe we can suggest improvements or whatever, that would be useful to you (and others) in terms of better (general) strategy development. I don't think you have anything to lose under this senario and on the contrary a whole lot to gain.
 
BTW, just wanted to throw in a word about walk forward testing. A lot of posters seem to believe I just threw this thing live. That is far from the case, it has been running live on ib's sim account for a LONG time... and through all of this walk forward testing it was VERY profitable... the only thing that changed is I went from the IB sim account to ib real money... nothing else.. hard to explain what went wrong
 
You must log in or register to reply here.
Back
Top