Trading What I see !?
Details fyi:
Number of markets = 5486
Number of selections = 42824
Number of Backs = 21729
Number of Lays = 21095
Back profit = 1346.99
Lay profit = 1655.91
Strike Rate = 57 %
Implied Back Odds = 1.75
Implied Lay Odds = 2.33
Avg bet price = 7.82
Avg bet size = 4.63
Avg weighted odds = 5.81
Wins = 3132
Losses = 2354
Av Win = 2.71
Av Loss = -2.40
Av Win:Loss Ratio = 1.13
Biggest Win = 54.83
Biggest Loss = -71.77
Highest Hi = 2871.22
Lowest Lo = 0.00
Biggest Drawdown = -130.50
Longest Drawdown = 3 days
Longest Time to NewHi = 6 days
Longest win streak = 15
Longest lose streak = 9
Total profit = 2829.71
Turn Over % = 0.49 %
Commission = 169.74 (6.00 %)
So last 7 days:
I've cut the bot back now, stopped 1 of the two strategies that definitely was performing poorly since new year, but still piss poor...
I'm giving it another day or so then moving to paper trading...
I'm actually thinking my strategy is possibly vulnerable to market change, as it works based on getting on value prices early on in the greyhound market 5-10mins out, when liquidity is low, but value is there, or at least it was last year! I reckon given i'm playing in low liquidity areas, it's a lot more vulnerable to some new or changing players simply blowing my edge... what do you experts think?
Seems I may have missed the boat on this one and go in at the top!
I'm giving it another day or so then moving to paper trading...
I'm actually thinking my strategy is possibly vulnerable to market change, as it works based on getting on value prices early on in the greyhound market 5-10mins out, when liquidity is low, but value is there, or at least it was last year! I reckon given i'm playing in low liquidity areas, it's a lot more vulnerable to some new or changing players simply blowing my edge... what do you experts think?
Seems I may have missed the boat on this one and go in at the top!
You do not have the required permissions to view the files attached to this post.
-
ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
Low liquidity or high volatility should only mean you can't get on or can't get enough on at your target price.goat68 wrote: ↑Tue Feb 08, 2022 10:00 pmSo last 7 days:
Screenshot 2022-02-08 at 21.53.45.png
I've cut the bot back now, stopped 1 of the two strategies that definitely was performing poorly since new year, but still piss poor...
I'm giving it another day or so then moving to paper trading...
I'm actually thinking my strategy is possibly vulnerable to market change, as it works based on getting on value prices early on in the greyhound market 5-10mins out, when liquidity is low, but value is there, or at least it was last year! I reckon given i'm playing in low liquidity areas, it's a lot more vulnerable to some new or changing players simply blowing my edge... what do you experts think?
Seems I may have missed the boat on this one and go in at the top!
If I saw an edge in my backtest and didn't see one when I traded it, I'd be looking at what I'm doing wrong rather than thinking about market changes. Afterall the 100% certain change is me getting involved, a market change is less than 100% certain. So I'd check....
1.Am I 100% sure I'm not backfitting? I'd then run maybe 10 random samples of 33% of my data each and check for correllation.
2.Are there any differences in execution between live and test? Check assumed latency vs actual, check the #bets placed, check % of bets matched and % of amounts matched to see if my testing was over optimistic.
3. That's enough to be getting on with.
Remember you can't retest the period you traded to check that test and live are the same, because you're present in that test data. You can try to force your test-self to get ahead of your live-self by cutting the latency but it's so inaccurate I woundn't bother trying that. All you can do is make sure the execution metrics are similar in your backtest (the data without you in it) and live, eg. seeing as many opportunities and getting similar amounts on, in a similar time, and getting matched in a similar time, at a similar rate.
When all that totally checks out then maybe it's a change to the market, but until you've proved the error isn't yours then that's pure speculation and an abdication of responsibility.
Thanks for the details as ever Shaun, appreciated.ShaunWhite wrote: ↑Tue Feb 08, 2022 10:36 pmLow liquidity or high volatility should only mean you can't get on or can't get enough on at your target price.goat68 wrote: ↑Tue Feb 08, 2022 10:00 pmSo last 7 days:
Screenshot 2022-02-08 at 21.53.45.png
I've cut the bot back now, stopped 1 of the two strategies that definitely was performing poorly since new year, but still piss poor...
I'm giving it another day or so then moving to paper trading...
I'm actually thinking my strategy is possibly vulnerable to market change, as it works based on getting on value prices early on in the greyhound market 5-10mins out, when liquidity is low, but value is there, or at least it was last year! I reckon given i'm playing in low liquidity areas, it's a lot more vulnerable to some new or changing players simply blowing my edge... what do you experts think?
Seems I may have missed the boat on this one and go in at the top!
If I saw an edge in my backtest and didn't see one when I traded it, I'd be looking at what I'm doing wrong rather than thinking about market changes. Afterall the 100% certain change is me getting involved, a market change is less than 100% certain. So I'd check....
1.Am I 100% sure I'm not backfitting? I'd then run maybe 10 random samples of 33% of my data each and check for correllation.
2.Are there any differences in execution between live and test? Check assumed latency vs actual, check the #bets placed, check % of bets matched and % of amounts matched to see if my testing was over optimistic.
3. That's enough to be getting on with.
Remember you can't retest the period you traded to check that test and live are the same, because you're present in that test data. You can try to force your test-self to get ahead of your live-self by cutting the latency but it's so inaccurate I woundn't bother trying that. All you can do is make sure the execution metrics are similar in your backtest (the data without you in it) and live, eg. seeing as many opportunities and getting similar amounts on, in a similar time, and getting matched in a similar time, at a similar rate.
When all that totally checks out then maybe it's a change to the market, but until you've proved the error isn't yours then that's pure speculation and an abdication of responsibility.
Im feeling a bit exhausted, so can't face all that at the moment, but saying that im reasonably sure ive not backfitted, also backtest of same live period does show strong correlation but as you say im in the market but i don't think that is significant as i dont believe my £4 bets move the price much...but don't know...
Tell you what ill DM you my complete strategy and perceived edge, as I think as someone else said I need experienced help...just to get your honest opinion on my mad ideas?!
Cheers
-
ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
It's a tricky week's work for you if done properly so no rush. And you might have to tweak your core stuff too to log some additional info. Project mngmt hat....i'd allow say 20hrs but end up with something reusable whenever I needed old test vs new live comparisons. Or for old live vs new live comparisons and daily performance/execution stats. Then in the future you can see if strategy failure is down to %matched etc or the queue being bigger than usual on average when your bets arrive or whatever. That might be cured by asking for less margin (a tick lower) or using a smaller stake....and if a strategy improves unexpectedly and the matched% has gone sky high you could ask for more margin or stake beyond what previously maxed it out. Loads of info to be had.
Right now you don't know what's changed, but something has and you'd look to your stats first for a clue because raw price data doesn't give many clues. My hunch is that you're not getting as much matched as the test does, and when that happens it's always the losers that match more more often as the price goes through you so a small change can have a fairly big impact.
Be 100% sure, you know how.
It's not just about moving the market it's more to do with position in queue. Your £4 live bet is in the data, your test £4 therefore either gets to the queue ahead of it or behind it. Both scenarios affect the amount the backtest needs to see matched before your test order gets matched. Ditto taking prices, test £4 might take an offer the live £4 misses or visa versa. Maybe not a big deal where 1000s are matched but an illiquid early dog might only have £40 or even just the £4 matched.
I don't think that would help much, frankly I don't have the time to configure a strategy in my code to produce the results I'd want to look at. You'll learn a lot anyway by examining the difference between live, different live sessions, and tests by things other than £. It'll be a quite a nice descrete side project for a week or two as a break from looking at dogs and you'll end up with a handy utility and some more knowledge, for free!
So i've done some of the above, and my reasoning still concludes the market has shifted...ShaunWhite wrote: ↑Wed Feb 09, 2022 1:09 amIt's a tricky week's work for you if done properly so no rush. And you might have to tweak your core stuff too to log some additional info. Project mngmt hat....i'd allow say 20hrs but end up with something reusable whenever I needed old test vs new live comparisons. Or for old live vs new live comparisons and daily performance/execution stats. Then in the future you can see if strategy failure is down to %matched etc or the queue being bigger than usual on average when your bets arrive or whatever. That might be cured by asking for less margin (a tick lower) or using a smaller stake....and if a strategy improves unexpectedly and the matched% has gone sky high you could ask for more margin or stake beyond what previously maxed it out. Loads of info to be had.
Right now you don't know what's changed, but something has and you'd look to your stats first for a clue because raw price data doesn't give many clues. My hunch is that you're not getting as much matched as the test does, and when that happens it's always the losers that match more more often as the price goes through you so a small change can have a fairly big impact.
Be 100% sure, you know how.It's not just about moving the market it's more to do with position in queue. Your £4 live bet is in the data, your test £4 therefore either gets to the queue ahead of it or behind it. Both scenarios affect the amount the backtest needs to see matched before your test order gets matched. Ditto taking prices, test £4 might take an offer the live £4 misses or visa versa. Maybe not a big deal where 1000s are matched but an illiquid early dog might only have £40 or even just the £4 matched.I don't think that would help much, frankly I don't have the time to configure a strategy in my code to produce the results I'd want to look at. You'll learn a lot anyway by examining the difference between live, different live sessions, and tests by things other than £. It'll be a quite a nice descrete side project for a week or two as a break from looking at dogs and you'll end up with a handy utility and some more knowledge, for free!
Here is 6 random 3rds from 2month data from last year, shows reasonable correlation: I have also just done some backtest v live comparison from the last 2days, and it shows very close match % and total, with 21.8% vs 22.2%. Also here is a backtest of the last 2days along with the actual live as well, as you can see fairly close: (ignore BSP y axis label, that's actually PnL)
You do not have the required permissions to view the files attached to this post.
Darn it! I think i've discovered a flaw/misstake, some of my backtesting was not measuring perfect hedge and I was measuring backtest performance and making adjustments based on incorrect data, or at least not ideal/neutral results. I'm now doing the september->jan backtest again with some new adjustments and correct "best hedge" result recording...ShaunWhite wrote: ↑Wed Feb 09, 2022 1:09 amIt's a tricky week's work for you if done properly so no rush. And you might have to tweak your core stuff too to log some additional info. Project mngmt hat....i'd allow say 20hrs but end up with something reusable whenever I needed old test vs new live comparisons. Or for old live vs new live comparisons and daily performance/execution stats. Then in the future you can see if strategy failure is down to %matched etc or the queue being bigger than usual on average when your bets arrive or whatever. That might be cured by asking for less margin (a tick lower) or using a smaller stake....and if a strategy improves unexpectedly and the matched% has gone sky high you could ask for more margin or stake beyond what previously maxed it out. Loads of info to be had.
Right now you don't know what's changed, but something has and you'd look to your stats first for a clue because raw price data doesn't give many clues. My hunch is that you're not getting as much matched as the test does, and when that happens it's always the losers that match more more often as the price goes through you so a small change can have a fairly big impact.
Be 100% sure, you know how.It's not just about moving the market it's more to do with position in queue. Your £4 live bet is in the data, your test £4 therefore either gets to the queue ahead of it or behind it. Both scenarios affect the amount the backtest needs to see matched before your test order gets matched. Ditto taking prices, test £4 might take an offer the live £4 misses or visa versa. Maybe not a big deal where 1000s are matched but an illiquid early dog might only have £40 or even just the £4 matched.I don't think that would help much, frankly I don't have the time to configure a strategy in my code to produce the results I'd want to look at. You'll learn a lot anyway by examining the difference between live, different live sessions, and tests by things other than £. It'll be a quite a nice descrete side project for a week or two as a break from looking at dogs and you'll end up with a handy utility and some more knowledge, for free!
-
ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
What was it for the backtest before you went live, say in sept? That's what we're looking for, differences between backtests that were profitable and the live execution which might not be. That stat from your retest of your 'used' data vs live seems quite good, a 1.8% difference albeit a very tiny sample. So now you've got an expectation that anything you've backtested should produce the same figures when you take it live, within that margin. So if a backtest was showing say 26% matched when you go live you'd be looking for that same (+/- about 2%).
The 6 sample chart looks good, if that's for 5 months (it's hard to see when you don't label you x-asis) then it looks like there could be spells of a couple of months at a time that don't look too good so maybe it's too early to draw any conclusions from your live yet? Those long period charts look good but when you live through each day in realtime you realise how long the downs and flatlines can feel.
-
ShaunWhite
- Posts: 9731
- Joined: Sat Sep 03, 2016 3:42 am
More haste less speed Grasshopper.
You do not have the required permissions to view the files attached to this post.
So the strategy is shit then?