The Bot Farm

Bitter experience has long taught me to use stop losses to limit my overall exposure for a given trade or gamble.  In most forex Expert Advisors the stop loss gets set automatically when the trade is entered.  Casino and horseracing bots have to stop when they run out of whatever small bankroll I have given them to use.  The main poker room I use has a seductive setting for automatic rebuys and top-ups which can drain your account dry overnight if your bot goes a little berserk.  Normally I don’t maintain enough in the account for this to be a major worry

Call it greed or optimism, but at the other end of the equation I have always resisted putting a profit target into my systems.  If asked what my target gain is I would be tempted to say “to the moon and back”.   At the back of my mind I always have the old trading adage “cut your losses, but let your profits run”.  The trailing stop loss is intuitively the most appealing approach to me, where the upside is unlimited, but only a limited amount of backsliding will be tolerated before the trade or session is terminated.  The problem with this is that it can cause you to lose out on large gains if the session would have recovered to above the exit point. 

Something like this consideration inspired the programmer of the Amoeba bot to implement a ”collar” system, where the bot would sit out and carry on trading in shadow mode only when a certain slipback from the local maximum profit was suffered, but would start trading live again when the session recovered.  The collared progression is seen to good effect in this graph, where the yellow line represents the actual profit curve achieved on the basis of the raw results, shown in white.   Had these results been consistently repeatable, of course, I would be writing this blog from a far warmer place.

When I used to frequent the Betfair Exchange Games forum, there used to be many arguments around the validity of “stop at a winner”, or “stop at a daily target” systems.  I was always convinced by the arguments that said if something has a winning edge you should be able to leave it running 24/7/365 and only worry about how to spend all the money you would be making.  However I have seen so many systems which run well over 100,000 games, spins or hands, only to descend an equivalent slope over the next period, that I have lost my mental resistance to snatching a few lucky breaks from the jaws of probability.  The ratchet effect has been on my mind a lot – this is based on the idea of the ratchet screwdriver, where you can twist it in equal amounts in both directions, but because its effective range of motion is restricted in one direction the result is unidirectional drive, with very little backsliding.  This might seem at first sight to equate just to stop losses, but actually the “pause for breath” is equally important, at the end of the effective movement, equating to taking profits and starting again.  

So far, so philosophical.  If you want something more grounded in practicalities, take the case of Rush poker, which has occupied so much of my recent waking time.  It is perfectly possible to construct a shotgun approach to blitz certain levels of this game with high degrees of aggression and risk taking, whether you are playing manually or with a bot.  Because most of your opponents are human, they can often be bluffed by such an approach into laying down what they fear may be the second-best hand, but which actually has you beat.  Such an approach can double or treble your buy-in very quickly.  But just as a wild west gunslinger would always run into someone faster, you or your bot will eventually run up against someone with the best possible hand and a bigger stack, who will end up collecting every cent you have on the table.  I have such a bot running at the moment using just such a shotgun profile.  Last time I looked it had run its buy-in up six-fold and I had to close it down and start up again from scratch for fear of it busting out.  Which it will inevitably do.  I saw it call an all-in with QQ after an A had come on the flop (it lost), call another with JJ which won against TT and – scariest of all - it called an all-in with nothing but overcards of AQ which won against a bluff of AJ.  So it will definitely crash and burn – I just need to make sure that when it does so it doesn’t plunge from too high a perch.   So – profit targets suddenly become a more attractive proposition.

As always, there is a software solution.  On top of the bot itself, which provides the framework, and the profile, which is designed to dominate the table, there are also cash hoppers, which will cause the bot to sit out when a certain stacksize has been achieved, then sit back into the game immediately with the minimum buyin.  I have bought two such items, but unfortunately can’t get them to work with the Windows Server 2003 operating system used by my Virtual Private Server.  So, until someone can help me to get them to work or suggest another approach, I’m going to sleep fitfully, worrying about what my little poker monkey is getting up to with my bankroll.