Trading What I see !?

[goat68]

Reality returns... September so far has just about wiped August !
Screenshot 2021-09-02 at 19.06.57.png
I'm always amazed how the handful of people can make 1000s per month from Betfair, as it just seems so impossible

1. What happened to make those two days so different to your backtests!?

With the tech you've got there really shouldn't be any suprises unless you're taking shortcuts. I know some people have told you stages they think aren't useful but they have years of experience to draw on. I'm fairly experienced but I still don't shortcut the methodology because I dont' like suprises like suddenly losing a load of cash. Never gamble unless you have all the information you can get. How much data do you have now and how much have people recommended you need?

2. You may still be too focused on the small picture which is understandable. As some sort of guide, a chart from the real world.

My wierd Aug....didn't even breakeven until the 15th. Randomness is a curious bedfellow you just have to get used to.
Screenshot_19.jpg

So I didn't say I didn't expect this sort of thing... just expressing the "reality" and frustration when it happens... so here's some backtest detail for "dogsbot": 3 month backtest May->Aug
Number of markets = 1526
Number of selections = 1682
Number of Backs = 675
Number of Lays = 1007
Back profit = 674.54
Lay profit = 595.98
Strike Rate = 58 %
Avg bet price = 3.53
Wins = 892
Losses = 634
Av Win = 6.52
Av Loss = -7.35
Biggest Win = 17.19
Biggest Loss = -17.34
Highest Hi = 1199.37
Lowest Lo = 0.00
Biggest Drawdown = -162.29
Longest win streak = 11
Longest lose streak = 6
Total profit = 1151.86

and here's the chart:

Screenshot 2021-09-02 at 20.18.07.png

[PapaShango]

Bad luck with the drawdown, goat68. I hope your graph starts going up again soon.

The big question is though, how much did ShaunWhite make this August? Was it £500 ? £5k ? £50k?!

You forgot to label the y axis on your graph again , Shaun !

[Black Ladder]

If you have a reasonable but consistent measured strike rate maybe test a fibonacci recovery plan with sub minimum stakes in live mode. With greyhounds I would also consider using another provider to broaden your expectations, and subsequent returns.

[ShaunWhite]

The big question is though, how much did ShaunWhite make this August? Was it £500 ? £5k ? £50k?!

50k you're joking right.
No I'm a small player, steady money but I don't work hard enough to make big numbers. tbh getting somewhere took me 2+yrs of 80hr weeks and no days off which almost buggered me, and because time has always been more important than money to me I cut down to about 35hrs/wk when it got to being enough. No point being coy and it's not exactly enough to brag about, Aug was 2½ but I usually expect about 3. £100 a day, so not that exciting really but the freedom makes me feel richer than when I used earn multiples of that doing more demanding jobs back in the day.

[ShaunWhite]

so here's some backtest detail for "dogsbot": 3 month backtest May->Aug

Good numbers on the surface of it but how certain are you it's not backfitted ?

I know you're dismissive of it but I'd love to see that divided up into random samples, it probably wouldn't take you long and it would give you some more reassurance. Then you can do an Excel CORREL(set1,set2) on the two sets of results (aka a rough and ready Pearson and hope it's >0.5ish). You could even go one step further and randomise multiple times (aka whacking the Excel F9 key with vba), which can give you a Monte Carlo of the correlation coefficients.

You'll see that as OTT I'm sure but a couple of evenings work and you've have another re-usable utility in your analysis suite that you can throw results into anytime and get that extra bit of knowledge in a few mins.

I just put this together as an example by backfitting a profitable dog strategy and divided it into 2 samples (about 500 races in each batch). I know you might think that a biggish sample means the results must be good but look at the sample 1 and sample 2 lines. The red line looks great, but there's so little correlation I wouldn't spent money on it. The two samples actually have a slight negative correlation, it might last a month or two but I wouldn't gamble on that.

Screenshot_20.jpg

I ran it forward another 5,000 races, I didn't want to spend ages so the scale looks odd but I got this. It made money for a while but then lost it, and it's well below where I might have started if I'd taken the first 1,000 at face value.

Screenshot_21.jpg

I'm not being a doubting Thomas I'm just trying to save you from the soul destroying false dawns. Loads of people find something that works for a while then say the strategy has died, it's maybe because there wasn't a strategy in the first place and winning was randomness? Most people would kill for the tech you've got and the ability to know what you can know.

...or maybe I do all that because losing isn't an option, I know it's a hobby for you and you're just trying and make a few quid, and you can afford to lose a wedge too so I shouldn't suck all the fun out of it for you.

mmmm Life's too short to do posts like that really. And you know all this anyway, so hopefully someone else might find it useful even if you don't.

[PapaShango]

Well done Shaun. £2.5k is not to be sniffed at.

Especially as it was being made whilst you were curled up on your sofa watching countdown!

Interesting seeing your drawdowns and variance in action though. If that was me, I’d have been panicking by mid August , good to see that your bot had the discipline to keep grinding out its edge and not do anything stupid !

[Anbell]

GOAT - here's a simulator that might help
https://valuebettingblog.com/advanced-b ... alculator/

[goat68]

GOAT - here's a simulator that might help
https://valuebettingblog.com/advanced-b ... alculator/

Thanks that's a helpful tool.

I'm a bit dubious of MC stuff, just by it's nature of having very little input it can't really tell you much. The thing it will tell you is the more bets you do the more confidence and smaller % drawdown you will achieve. So basically it tells you the more bets you do the more confident you will be, which is sort of obvious!

It doesn't tell you if you've fitted your strategy, which is my biggest issue. I try and avoid fitting by using multiple sample sets, typically about 1month's worth each. However, what I find is market behaviour or my fitting changes from month to month. Which might actually be the case with "seasonality"...

[Anbell]

GOAT - here's a simulator that might help
https://valuebettingblog.com/advanced-b ... alculator/

Thanks that's a helpful tool.

I'm a bit dubious of MC stuff, just by it's nature of having very little input it can't really tell you much. The thing it will tell you is the more bets you do the more confidence and smaller % drawdown you will achieve. So basically it tells you the more bets you do the more confident you will be, which is sort of obvious!

It doesn't tell you if you've fitted your strategy, which is my biggest issue. I try and avoid fitting by using multiple sample sets, typically about 1month's worth each. However, what I find is market behaviour or my fitting changes from month to month. Which might actually be the case with "seasonality"...

I doubt that seasonality is your issue.

[goat68]

I've done an interesting test looking at my "hedge" bet value.
So I used a strategy i'm working on for AUS horses, and used hedging at post time. Applying the method across the broad selections, and then hedging showed a net loss, BUT my net on just the "hedge bets" was about £0.
Then applying filters on the signal to find a better edge, showed +EV on the "signal bets", but a significant loss on the "hedge bets" -£343.
However, I know that if I don't "hedge" the +EV "signal bets" end up with a net loss due to commission!

[ShaunWhite]

GOAT - here's a simulator that might help
https://valuebettingblog.com/advanced-b ... alculator/

Thanks that's a helpful tool.

I'm a bit dubious of MC stuff, just by it's nature of having very little input it can't really tell you much. The thing it will tell you is the more bets you do the more confidence and smaller % drawdown you will achieve. So basically it tells you the more bets you do the more confident you will be, which is sort of obvious!

It doesn't tell you if you've fitted your strategy, which is my biggest issue. I try and avoid fitting by using multiple sample sets, typically about 1month's worth each. However, what I find is market behaviour or my fitting changes from month to month. Which might actually be the case with "seasonality"...

Don't do month A then month B, randomising across it all.

MC is only intended to show you the expected range of results, ie how likely is your outcome. If you're dubious perhaps you've misunderstood what its for.

A month of data in each set? You know that's not enough so why do it?

[goat68]

GOAT - here's a simulator that might help
https://valuebettingblog.com/advanced-b ... alculator/

Thanks that's a helpful tool.

I'm a bit dubious of MC stuff, just by it's nature of having very little input it can't really tell you much. The thing it will tell you is the more bets you do the more confidence and smaller % drawdown you will achieve. So basically it tells you the more bets you do the more confident you will be, which is sort of obvious!

It doesn't tell you if you've fitted your strategy, which is my biggest issue. I try and avoid fitting by using multiple sample sets, typically about 1month's worth each. However, what I find is market behaviour or my fitting changes from month to month. Which might actually be the case with "seasonality"...

Don't do month A then month B, randomising across it all.

MC is only intended to show you the expected range of results, ie how likely is your outcome. If you're dubious perhaps you've misunderstood what its for.

A month of data in each set? You know that's not enough so why do it?

Struggling to find anything that works across 3 months on pre-horses just on plain calendars! don't have to worry about randomising yet!

[goat68]

I'm actually coming to the conclusion it's pretty darn hard pre-race horses to find anything consistent across 3months, behaviour seems to change from month to month, but it's obviously possible, Shaun does it!

[goat68]

GOAT - here's a simulator that might help
https://valuebettingblog.com/advanced-b ... alculator/

So there must be the answer to this somewhere on this forum... but for the purpose of this calculator what is the Yield formula for Lay bets? The above calulator uses the term "Volume"? So if I Lay £2 at 5.0, the "Volume" is £2 ? or is it the risk amount ie.£8 ?

[goat68]

It's frustrating these websites all seem to just view betting just as placing Back bets...which I guess it is! but make Yield tricky, and as such can't use the above calculator as I need to give it the Yield value it is expecting...>?