Why does scaling up make a winning bot fail?

[Fugazi]

I can't figure out why increasing the stake causes my previously profitable automated strategy to fail.

Initially, I thought the reason was obvious: poor value bets are getting matched while good value bets are ignored.

But if this were the case, the strategy should also be unprofitable at low stakes. Even with small bets, the good value bets remain untouched while the bad value bets get matched. So why is it profitable at low stakes but not at higher ones?

My alternative hypothesis is that it's about manipulation risk. When the bot places only £1 bets, it's not worthwhile for others to manipulate the market to exploit it, because they risk losing their own £1 without significant gain. However, if the bot starts placing £2 bets, manipulators might be more inclined to act since they can risk £1 for a chance to profit from the bot's £2 bet. The risk-reward ratio improves for the manipulator as the bot's stake increases.

But this doesn't explain why, even when I program the bot to wait a random, lengthy period between a bet being matched and placing the next one, the strategy still becomes unprofitable simply because the total stake exceeds the minimum bet.

Can someone help me make sense of this?

[jamesedwards]

I figure there's several reasons why expected value % drops inversely with stake. Here's a more obvious one:

The higher your stake the longer it takes to be fully matched, therefore the lower your expected % return.
Let's say in an extreme example, you want to back more goals in a football match with 20 mins to go and price at 2.0.
Backing £1 @ 2.0 you might fully match in 10 secs and still have 19m50s plus injury time to get the goal you need.
Backing £5,000 @ 2.0 you might take 3 mins to fully match so only have 17m plus injury time to get your goal.
You've held up the price for 2m50s longer than otherwise may have happened so have given away value. Your expected % return has declined by the average amount of probability of a goal being scored in those lost minutes.

Upon becoming part of any market your very presence shifts value from your side to the other. The greater your stake the greater the shift and therefore the greater the impact to your expected return%.

[Fugazi]

Yes, this would explain why even when I put a time gap between small bets rather than a larger stake I still have the same issue of it becoming unprofitable.

Optimising my bot settings to any real precision is too difficult, so I can't pinpoint where in the timeline my bot becomes unprofitable and cancel all bets after x period of time.

Frustrating as if I could optimise my bot I could probably increase its profit 10 fold

[weemac]

Which sport, and how long pre-event? It makes a big difference, even at small stakes.

[Fugazi]

Which sport, and how long pre-event? It makes a big difference, even at small stakes.

Horse racing 24 hours - 20 hours ish depends on volume and some other conditions

[Anbell]

I can't figure out why increasing the stake causes my previously profitable automated strategy to fail.

"Cause" is doing a lot of work there. How confident are you about it?

[Fugazi]

I can't figure out why increasing the stake causes my previously profitable automated strategy to fail.

"Cause" is doing a lot of work there. How confident are you about it?

In what way? Something changes when stake gets increased. Sample size and ROI are sufficient at £1 I'm confident it works. It operates in a very narrow window of conditions so doesn't take a lot to upset it.

Maybe its not direct, maybe as per the other comments something else happens when stake increases eg. The bet takes longer to match, and the longer it takes to match the more likely someone has waited for it to become bad value

[sionascaig]

Do your bigger stakes stick out in the market in some way or do they look innocuous?

And don't rule out variance - Euler posted something a month or so back about favs winning significantly more than expected...

[LinusP]

scaling is hard

https://robotjames.com/posts/scaling-is-hard/

[Anbell]

In what way?

There are a million things going on in the market. You state that increasing stakes *causes* your bot fail. Are you sure about that? Or might it be variance?

[Anbell]

(It is also true that scaling can be hard)

[Fugazi]

scaling is hard

https://robotjames.com/posts/scaling-is-hard/

I really like that.

"well, not easily.. cos everything is a lot of annoying parochial effort."

Sums up my problems with horizontal scaling and trying to optimise settings when i need a months or more forward testing every time I want to make a minor tweak

[jamesedwards]

This thread reminds me of my greyhound automation from a few years back.

Several thousand races here. This is when the penny really dropped with me on the impact of stake.

z15.PNG

[weemac]

This thread reminds me of my greyhound automation from a few years back.

Several thousand races here. This is when the penny really dropped with me on the impact of stake.

z15.PNG

And so sensitive to any and every bet are overnight racing markets, that even moving from £1 to £2 might be a significant issue. You can hide £1 on the ladder (to all of the people some of the time and some of the people all of the time), but £2 is so much more visible to everyone.

On reading the OP, I immediately guessed it would be racing markets a long time pre-off. I reckon the issue is (in)visibility.

[Fugazi]

This thread reminds me of my greyhound automation from a few years back.

Several thousand races here. This is when the penny really dropped with me on the impact of stake.

z15.PNG

And so sensitive to any and every bet are overnight racing markets, that even moving from £1 to £2 might be a significant issue. You can hide £1 on the ladder (to all of the people some of the time and some of the people all of the time), but £2 is so much more visible to everyone.

On reading the OP, I immediately guessed it would be racing markets a long time pre-off. I reckon the issue is (in)visibility.

Yep. I actually remember seeing James post that before. Couldn't remember who made the post or what it was about but I knew someone profitable had a very similar problem increasing stakes.

I think it is a visibility issue. When I raise the stakes it becomes obvious which is my bot, and then someone must set to work figuring it out. I made its entry rules as disguised as possible, but ultimately probably doesn't take long for someone to spot how they can make a small profit for the day abusing it