The BTC Horse Racing Thread
-
Cracking day
-
0_1676290841246_Pace Backs New.json
A new version of the pace backs taking into account what another user suggested. The thing i really like about the equity curve on this is there is a clear pattern. The equity curve is similar in both years, for example shows good growth over the summer but in both years October and November are flat.
-
@chris-osborne
Its working for me now. Just had about 10 mins where fairbot wasnt connecting and the website wouldnt load at all.
-
@jonathan-jones Fine for me on both website and Fairbot
-
Is betfair dead?
Cant log on either fairbot or the website. -
@adam
Things have improved massively and its still improving.
As i said its only a couple of the strategies that have this issue, Others are fine.And, i have Offered solutions to get over that limitation AND have commented on the fact they do make the results more reliable.
-
@nick-allan said in The BTC Horse Racing Thread:
@jonathan-jones interesting jonathan, keep us posted with results as i for one would be interested. Yes lay the beaten fav is struggyling of late, but not sure why. And yes the backtesting is impressive but in the live enviromentment over the last 2 months its not
All I can suggest is that you make sure you're testing what you think you are. The backtesting is telling you what actually happened, it's not trying to make any "impressive" predictions.
-
@joshua-varley said in The BTC Horse Racing Thread:
to
Yes it would. It would reduce the inconsistency but not eliminate it entirely.
But,
If you test that in the backtester entering at 10 seconds with the favourite conditions reduces profitability to 389 while removing the criteria and still entering 10 mins before start only reduces it to 429 (and, by moving the time you are only reducing the inconsistency, not eliminating it).Lay the beaten fav also has this problem. There are loads of other criteria you cant use either because of this.
-
@jonathan-jones interesting jonathan, keep us posted with results as i for one would be interested. Yes lay the beaten fav is struggyling of late, but not sure why. And yes the backtesting is impressive but in the live enviromentment over the last 2 months its not
-
@jonathan-jones said:
to see how misleading the software actually is
@jonathan-jones said:
Its results are not replicated in real trading.
@jonathan-jones said:
In reality you will bleed money while the software happlily tells you that you are making a profit.
@jonathan-jones said:
live trading should be a total match to the backtest.
@jonathan-jones said:
those results are a fiction
@jonathan-jones said:
I would rather something i can trust
We've been over this many, many times. The only "favourite" rule we have right now is BSP favourite. BSP is mentioned in the dropdown hint. BSP is mentioned in the help dialogue when you click the rule icon. BSP has been mentioned a thousand times whenever you've had a dig about it on this forum.
We are not trying to mislead anyone and this is not an indication that the software is unreliable. Something being missing from the software does not indicate that the data that IS present is wrong. You are comparing apples with oranges. If you want to know the exact favourite at some other timestamp other than when BSP is calculated, this is one of the few things the backtesting cannot do right now. Some people may be happy with an approximation in the last few minutes before the race, others may not be and may prefer to leave the rule out for now. Either one of those is fine.
Please learn the difference between "the software is wrong" and "there's a cool feature that's not in the software yet".
Are you going to add a better "Favourite" rule?
Yes. It is the main big update we are working on for the horse racing software. We are also doing the same update for all other stats that change over time during the lifetime of the market, for example traded amounts, going, number of runners, etc etc (there are about 20 different stats for which we're developing this feature). And obviously we will apply it to the other new software we're developing like the new version of the football software, etc.What's taking so long?
Right now, taking favourite as an example, we store 1 value: The BSP. We calculate the favourite based on the selection with the lowest value. Going forward, we need to (potentially) store a value every time a price snapshot is taken (once per second pre-race, once every 500ms in running). We need to do this for numerous metrics. We also need to re-write the strategy query logic to allow it to pick out the favourite value for any specific timestamp along that timeline. None of this is trivial. We need to think about the size of the data and the performance of the queries.Why didn't you put it in from day one?
Honestly? We just didn't think of it. We developed the framework for backtesting various values, of which the vast majority are fairly static (race type, course, etc). Then we went through all the data we were scraping and identified other values we thought would be interesting (going, traded volumes, etc etc) and applied the same logic. It was only when this was requested that we saw the difference, and the benefit of enhancing those stats, and began working on it. I guess you can add this to the list of 200+ other requests and great suggestions that have already made it into the software.Why didn't you psychically tap into my brain before I signed up and add everything I personally want to the software in advance?
My crystal ball was on the blink that day. I wholeheartedly apologise. If some feature is not in the software right now and you consider it a deal-breaker, please feel free to take some time away and sign back up when we let you know it's released. You are totally free to do so.I really, really hope that clears things up.
-
@joshua-varley @Jonathan-Jones this is what I had considered - the backrest obviously knows which horse the favourite is at the off, but when we place our bet at 10 mins before the off we don't know if that horse will go off as the favourite or not. So until we have some means of tying the results to real life could we not just ask the bot to select the favourite at eg 10secs before the official start time? Actual start time would be better but can't see how that is achievable.
-
@jonathan-jones this might be totally silly, but if you set the entry time on the c&d and bf strategies to much closer to the official time (obviously no point in doing it to the actual time, as doesn’t work for automation), would this not give a clearer indication of which horse is likely to be/ likely not to be fav? Obviously there’s going to be slight discrepancies as the actual off time is almost always slightly later than the actual, but it would surely be more accurate than 10 mins out?
Setting the entry time to 2 secs before the official off time and then again 2 secs In play (when the software is no longer looking into the future), shows no major difference in selection totals/SR%, meaning the 2 second before method was pretty damn close to reality.
Obviously placing a trade in-play is virtually out of the question, so bringing the entry time as close to the official off as possible, seems to me to be a way to overcome the issues you have raised?
Pretty new to all of this, so I could be completely barking up the wrong tree here haha…
-
Until the issue with favourite is resolved i cant see why not.
The results with them on backtest are obviously better but those results are a fiction. There simply isnt any possible way to know what the actual results in real trading with favourite would be, and its not a case of looking at the results without them and saying thats a worst case so reality would be something between. Thats not how it works. The reality could be you blow your bank by trading it with them. Point is, you simply dont know and the software as it stands wont tell you.
I would rather something i can trust (even if its far from optimal).
-
@jonathan-jones Hi mate, so you're simply saying with the "C&D winner" strategy, we should untick the "favourite", "2nd fave" and "3rd fave" boxes and re-run it? And these results would be more reliable than the previous ones? They're both showing very similar stats in terms of P/L, strike rate, ROI etc...
Thanks.
-
This may of of interest regarding your experiment.
Today i re-downloaded alll the individual results for the strategies (i do have but am ignoring Pace backs for now as its obviously on a long term very bad patch).With the BackC&DWinner and LayBeatenFav as they are (with the favourite conditions, so backtested results cant be trusted) over the total perios (since March 21) you would make 1706.5 points profit.
Your Biggest loss on a day would be 35.85 points.
Your Biggest win on a day would be 52.51 points.
Your Average on a day would be 4.82 points.Your total drawdown would be 45 points which would of occured last week
With the BackC&DWinner and LayBeatenFav without the favourite conditions, over the total perios (since March 21) you would make 1630.64 points profit.
Your Biggest loss on a day would be 39.48 points.
Your Biggest win on a day would be 48.56 points.
Your Average on a day would be 4.6 points.Your total drawdown would be 56 points which would of occured back in March 21
-
Dont use back the C&D winner. Its results are not replicated in real trading. In reality you will bleed money while the software happlily tells you that you are making a profit. You can have a go at testing it going orward withoug the 'favourites' condition. The backtest says that should also be nicely profitable (just a little less so with a tad more variance) but it live trading should be a total match to the backtest.
-
I have just started a similar test (well, on hold after last week while i reevaluate). Last week had the biggest DD ever. Massively blew out the previous 2 year record. Though i dont think the software will show that.
The issue is the LAy Beaten Fav and back C&D winner. Both use 'favourite' as a creiteria and simple fact is you cant use it. It doesnt work. Its ok on the back the fav strat because that triggers in the last 10 seconds and looks at a low odds range but for the others its just crap. Last week i lost several points the software ignores because the top 3 just swap position several times in the last 10 mins. YOu could argue that it should even out over time (as i did pick up a couple of points from the inconsistency) but thats gambling, its not something thats verifiable, backtestable or could be replicated.
Interesting thing is both strategies are profitable withougt that condition (and the lay beaten fav actually seems to be better with recent berformance without it) so this morning i am going to re-download all the results and re-analyse both with the condition (to see how misleading the software actually is) and without to see if the DD is manageable running the strategies in a more reliable form.
-
@david-milligan said in The BTC Horse Racing Thread:
@john-folan do you lay to a liability on all your strategies that aren't short odds? Is this to bring the risk reward under control a bit more? If so what sort of staking have you been using?
I do yes and quite heavy depending on drawdown.
