Does anyone want to join me helping test a value betting / dutching strategy?

[zippus]

I was wondering if anyone on the forum would like to team up to test out and refine a value betting strategy I'm working on using dutching. The strategy itself involves comparing the top 2, 3 and 4 book values to the actual top 2, 3 and 4 win rates for specific venues and race types and then placing back or lay bets where the greatest value is to be found.

I've made good progress so far on data analysis and automation but could do with a fresh pair of eyes and also some split testing to see where value can be added. I'd obviously share the data and automation files with anyone interested. This is what the output looks like on Guardian.

Guardian_280821.png

The results so far are mixed, with the lay dutching strategy performing reasonably well but the back dutching performing poorly (see graph below).

Equity curve_280821.png

[ShaunWhite]

First question would be why do you think this works and why do you see it continuing?

When you mine price data you can always find edges by filtering, track, distance, what it had for breakfast etc but finding those that are statistically significant and/or intrinsically present are harder.

Either through in/out of sample tests, MC sims, or maths statistics, the first thing I suggest is to prove to yourself this isn't randomness.

[zippus]

First question would be why do you think this works and why do you see it continuing?

When you mine price data you can always find edges by filtering, track, distance, what it had for breakfast etc but finding those that are statistically significant and/or intrinsically present are harder.

Either through in/out of sample tests, MC sims, or maths statistics, the first thing I suggest is to prove to yourself this isn't randomness.

Thanks for the input Shaun.

Re the first question. It's currently at early stages and I don't know if it works yet. It just seems logical to me to seek to find value by comparing the actual probability of top 2, 3 and 4 horses winning with the market's view (which I take to be the book %).

I totally agree with you that proving it isn't randomness is key, which is why I'm looking to continue this for approx. 1,000 races and to then use split testing to seek to prove that the strategy is working.

[ShaunWhite]

I'm not sure my input was very useful

I've been thinking about this and how my approach had changed over the years. I, like you, used to look for traps, tracks, distances, prices, days etc etc and try to find which were winners. These days I look from the other end, at what bets would have been winners and then look for what connects them. It's a small difference in your angle of attack but can make a big difference although it's a bit hard to get your head round if you've always thought the other way.

As an extreme example: If you found that all 3rd favs at Swindon were backing +EV. Then you'd be limitted to only being able to back the 3rd fav at Swindon, and frankly savvy punters don't let that sort of thing persist for too long. Whereas, if you could establish that winners were dogs that shortened by 5 ticks in the last 20 seconds or saw bets of a certain volume around the time they're paraded, then you could potentially bet on a much larger number with all the benefits that gives you. And more importantly it could result in more selections per market.

Scale is so so important because being able to bet(trade) more than one selection per market gets you an instant commission discount.
Example again:
Two strategies both with an SR of 70% and both either win or lose £1.
Trade them in seperate races and you'll pay commission on 70% of your markets.
Trade them in the same markets and you'll only win on both and pay commission 49% of the time. The rest of the time either just one wins (+1-1=0), or neither do.
You're paying commission on 21% fewer markets, losses offset the wins where it matters, where BF draw a line across your bets and charge you commission. Whereas a loss in a seperate market does nothing to help offset the commission.

Example again, my Aus horse bot makes ok money and bets on about 4 horses per market. If it was 1 horse per market then I'd be down about 30 grand this year already. Aus comm is much higher so it has more effect, but the rest of my stuff is similar. Widening your filter to include more losers can actually result in a bottom line gain if those additional selection are in the same markets as your others.

Bit of a ramble there zippus but maybe one or two things to think about.

[zippus]

No it was helpful input and I really appreciate you taking the time to respond.

Great news re your Aussie bot. That’s a decent return! I’ll give some thought to the points you’ve raised.

One of the things I’ve noticed over the small test sample is that the back dutches are poor on handicaps. I’d have expected this to be honest but I can’t get my head around the fact I’m using actual win rates, so I should get a return and a decent price, albeit less frequently than with races such as stakes. Lots to think about…

[ShaunWhite]

I said I made "OK money", which would be a huge loss without aggregation. It's decent but not exclaimation mark decent

[Trader Pat]

I'm not sure my input was very useful

I've been thinking about this and how my approach had changed over the years. I, like you, used to look for traps, tracks, distances, prices, days etc etc and try to find which were winners. These days I look from the other end, at what bets would have been winners and then look for what connects them. It's a small difference in your angle of attack but can make a big difference although it's a bit hard to get your head round if you've always thought the other way.

As an extreme example: If you found that all 3rd favs at Swindon were backing +EV. Then you'd be limitted to only being able to back the 3rd fav at Swindon, and frankly savvy punters don't let that sort of thing persist for too long. Whereas, if you could establish that winners were dogs that shortened by 5 ticks in the last 20 seconds or saw bets of a certain volume around the time they're paraded, then you could potentially bet on a much larger number with all the benefits that gives you. And more importantly it could result in more selections per market.

Scale is so so important because being able to bet(trade) more than one selection per market gets you an instant commission discount.
Example again:
Two strategies both with an SR of 70% and both either win or lose £1.
Trade them in seperate races and you'll pay commission on 70% of your markets.
Trade them in the same markets and you'll only win on both and pay commission 49% of the time. The rest of the time either just one wins (+1-1=0), or neither do.
You're paying commission on 21% fewer markets, losses offset the wins where it matters, where BF draw a line across your bets and charge you commission. Whereas a loss in a seperate market does nothing to help offset the commission.

Example again, my Aus horse bot makes ok money and bets on about 4 horses per market. If it was 1 horse per market then I'd be down about 30 grand this year already. Aus comm is much higher so it has more effect, but the rest of my stuff is similar. Widening your filter to include more losers can actually result in a bottom line gain if those additional selection are in the same markets as your others.

Bit of a ramble there zippus but maybe one or two things to think about.

Good post Shaun.

If there was a 'useful posts' thread that would be in it.

[napshnap]

...

[napshnap]

Ts, sorry for a little off topic, but I'm eager to ask Shaun some questions.

I've been thinking about this and how my approach had changed over the years. I, like you, used to look for traps, tracks, distances, prices, days etc etc and try to find which were winners. These days I look from the other end, at what bets would have been winners and then look for what connects them. It's a small difference in your angle of attack but can make a big difference although it's a bit hard to get your head round if you've always thought the other way.

Shaun, you mean transition from Fundamental -> Technical to Technical -> Fundamental, right?

Whereas, if you could establish that winners were dogs that shortened by 5 ticks in the last 20 seconds or saw bets of a certain volume around the time they're paraded, then you could potentially bet on a much larger number with all the benefits that gives you.

Shaun, how do you see the Market, what is it consist of, for example - we calc VWAP (excluding obvious outliers) for the whole time period 10s or 20s or 3m (or whatever) before start and than we compare this VWAP with a LTP or VWAP or top back (or whatever) for final time period, final 10s, 20s or 3m. Does it make sense to you, can we say that this pre-pre-start VWAP is a different part of the Market, maybe less professional than this final pre-start VWAP which is a pro's (bookies, pro gamblers, anyone who awaits the Market to form and accumulate enough iquidity) Final Word? What do you think about it?


Don't you look at Fundamentals at all for Aus bot?

[ShaunWhite]

I'll reply naps but wanna do it properly so probably later.

[napshnap]

I'll reply naps but wanna do it properly so probably later.

Of course, Shaun, take your time.

[ShaunWhite]

Shaun, you mean transition from Fundamental -> Technical to Technical -> Fundamental, right?

Not quite as radical as that. More like I'd previoulsy sort/group by distance or track or whatever to see which were good. Now I'd sort by results and try to find the common thread. Both sort of blur admittedly but it was more a change to the way I'm thinking. 'Winning' tracks have losers and 'losing' tracks have winners so what on earth is it that makes winners identifiable?!.....or lets say 'more identifiable' than which track they've turned up at. The main aim really is to make any animal anywhere a potential target rather than making that list shorter and shorter by filtering on a fundamental*

* Terminology: We all sort of make our own up but for me...
Technicals = things which markets have in common, they all have prices,times,volumes, moves, a time when it'll close etc etc.
Fundamentals = things which make markets different. Distances, class, form, track, jockeys etc.

My view almost certainly comes from not really being interested in the sport. I sort of see a dog market and a horse market being like most other markets that go up and down according to supply and demand. The fact it's wheat or corn doesn't really bother me. BUT.... I also know people who make my results look like pocket change who really get into the differences, so as always there's more than one way to do this.

Shaun, how do you see the Market, what is it consist of, for example - we calc VWAP (excluding obvious outliers) for the whole time period 10s or 20s or 3m (or whatever) before start and than we compare this VWAP with a LTP or VWAP or top back (or whatever) for final time period, final 10s, 20s or 3m. Does it make sense to you, can we say that this pre-pre-start VWAP is a different part of the Market, maybe less professional than this final pre-start VWAP which is a pro's (bookies, pro gamblers, anyone who awaits the Market to form and accumulate enough iquidity) Final Word? What do you think about it?

Again just my take but I'd say you're right, there's different phases. Earlybirds, after the prev race finishes, near the off time, between scheduled start and actual start, in-running.... I actually think there's a small 'thing' around blue-collar lunchtime too when people might have a bet or sneak to the bookies for half and hour (I always did anyway ) I can't say that one period necessarily has much bearing on the next, but within those periods they're similar. Eg what works super late on one will maybe work super late on another. Does early affect late, probably not.

Don't you look at Fundamentals at all for Aus bot?

Nope. I could but I'm so pathalogically bothered by backfitting, and so concerned that any edge that takes a year to find really has to last >1yr to pay it's way that I'm almost allergic to parameters, conditions, filters etc. If you only have one signal then there's only one thing that can go wrong, and one thing to tweak to fix it. Even 3 or 4 ifs, buts and maybes creates so many perms you've no idea what's doing what.

I'm not advocating anyone changes what they do but we get a lot of talk on here about excluding losers to find winners, filtering, handicaps/non-handicaps etc that it's not always obvious there's also a more broad brush approach. Almost certainly less £/animal but you can be betting on 800 a day rather than 20. Similar bottomline but your intel accumulates like nobody's business and you don't need months to find out what's working.

[ShaunWhite]

.....sorry naps re-reading that it's a "yes", it was from fund to tech but also different in that I wasn't looking at the tech to find a profit. I was seeing what profitted then looked for why. small diff but it resulted in also only considering the techs because I wanted to deliberately rule-out the fundamentals.

lol i'm confusing myself now

[zippus]

Thanks for the contributions. I particularly like Shaun's comment:

"Now I'd sort by results and try to find the common thread."

I guess I'm at the stage of being ok at finding WHERE there are issues but have little clue as to the WHY. I think book value and the movement of money on the top 4 (or however many 80% of the book repsesents) is key. I'm just loathed to follow the money all the time because that will just mean backing strong favs and that doesn't work. I'm trying to find where the market view is wrong...

[ShaunWhite]

I'm trying to find where the market view is wrong...

Value betting or looking for moves? (different things just for the purposes of this convo)

How far are you trying to look ahead? It's easier to figure out where it's wrong in relation to where it will be in 5s than where it's going to be in 30s (obviously). tbh if you can reliably say where it'll be in 2s you'd be a rich man. There's a lot of 2s chunks in 5mins.