10 000 bets into BowTied BetPicks - An analysis, part 1

A license to print?

Jun 03, 2024

Do you have a friend who likes betting and would like to sharpen his/her action? Help them on their path by sharing the BowTiedBettor Substack. Win and help win!

Share BowTiedBettor

Back in 2023 we launched a couple of bet feeds/streams [BowTied BetPicks] to provide our audience with a reasonable way of *actually* beating the betting markets.

Since launch we’ve now collected ~10 000 data points from both main services, the UK Horse Racing Channel & the US Sports Channel. With samples of decent size at hand, the time has come for an extensive analysis of the service.

Today we’ll share the analysis for the UK Horse Racing Channel & then next week we’ll be back with a part 2 that contains a similar investigation of the US Sports Channel.

In case you’re new to BowTied BetPicks and don’t have a clue what this is all about, please refer to the dedicated website.

Note I: As time progresses & more data keeps coming in we continue to adjust the current framework to further optimize the way the Bots operate and learn from the different markets & market movements. Great strategies always a result of trial & error/constant refinement.

Note II: Uninformed critics will claim ‘if a betting market edge actually existed you wouldn’t be selling it’. NGMI. Subscribers have most likely noticed the lack of Betfair trades & parimutuel/tote recommendations in our services. It’s all soft book action. The reason? ***Sharing profitable soft book action with you doesn’t cut into our personal earnings potential, but rather enhances it [subscription payments]. If it did, we’d turn it off/increase price/restrict access. Simple as.***

Below we’ll share several different ways of looking at the historical data to help you take max advantage of the alerts/feeds. To aid your understanding, most of the information will be presented in [hopefully intuitive] plots/graphs. If anything’s unclear, please let us know & we’ll clarify in the comments section.

If you’ve, after having read the complete analysis, still got unanswered questions you’d be curious to look into for yourself, feel free to request access to the full dataset/s by emailing us at betpicks@bowtiedbettor.com.

UK Racing Channel

Let’s jump straight into it!

Totals

Bets: 8 790.

Average number of bets per day: 45.

P&L [assuming flat staking]: +1361 units.

ROI: +15.4 %.

CLV [against Betfair starting price]: ~17 %.

Results assume you shop around & bet the selection with the bookie offering the highest price.

Note: Rule 4 deductions [cut into posted profits] are NOT included in the computations. Neither are improvements from “best odds guaranteed”-offers [yield higher odds on drifters & thus improve results].

By bookmaker

A bet is deemed ‘available’ at a bookmaker if the price at that bookie is within 2.5 % distance of the max price for all bookmakers for a given notification [essentially translates into it being offered at the same price or higher than the rest of the bookies for a horse]. This is a sensible way of approaching the problem of whether to bet or stay away from a selection at a particular bookie since there’s generally *some* info in the prices the rest of the market puts up. If the Bot says there’s 10 % EV on a horse where Bet365 is top of the market it’s much more likely to be a correct observation than if the Bot identifies an ‘equivalent’ 10 % EV on a selection where they’re sitting 20 % below the rest of the bookmakers.

Make sure to use this fact/insight to your advantage when going live with this strategy [or any other…].

Bet365

Stats for bet365:

Number of bets: 5364 [58 % of all bets available here]
Total P&L: 965 units
Expected P&L: 960 units
Average CLV: 17.91 %
ROI: 18.0 %

The closing line value measures how much value we’re taking against the best available approximation of the ‘true’ starting price of a horse. If the average of this distribution is above zero, we’ll soon see ourselves collecting profits from the markets…

Unibet

Stats for unibet:

Number of bets: 1565 [17 % of all bets available here]
Total P&L: 79 units
Expected P&L: 207 units
Average CLV: 13.29 %
ROI: 5.11 %

888sport

Stats for 888sport:

Number of bets: 2410 [26 % of all bets available here]
Total P&L: 305 units
Expected P&L: 340 units
Average CLV: 14.14 %
ROI: 12.67 %

Kambi

Stats for kambi:

Number of bets: 2406 [26 % of all bets available here]
Total P&L: 295 units
Expected P&L: 375 units
Average CLV: 15.62 %
ROI: 12.29 %

By parameter…

Time until post time

Does an illiquid, early BF market contain informative action & not only pure noise? It sure does appear so!

Note: CLV [closing line value] is measured against Betfair Starting Prices [BSP], which is generally considered to be the best available proxy for ‘true’ win probabilities. If your strategy consistently beats BSP, it’ll [almost ;)] certainly make you money.

Average CLV by time until post:
time_bin
<5         0.152576
5-60       0.147325
60-360     0.148900
360-600    0.186298
>600       0.203081

P&L and ROI by time until post:
             P&L      ROI   # of Bets
time_bin                          
<5        199.51  0.169507  1177.0
5-60       43.11  0.057327   752.0
60-360    258.44  0.084651  3053.0
360-600   568.30  0.188054  3022.0
>600      291.78  0.371694   785.0

Odds

Average CLV [%] by Odds:
odds_bin
<4      13.508262
4-6     18.583440
6-8     21.043741
8-12    21.214458
>12     25.451369

P&L and ROI by Odds:
             P&L       ROI    Bets
odds_bin                          
<4         79.20  0.117682   673.0
4-6       225.94  0.153910  1468.0
6-8       190.00  0.099685  1906.0
8-12      334.00  0.115691  2887.0
>12       532.00  0.286792  1855.0

Matched volume on the specific race [inversely correlated to time until post]

CLV against matched volume at the Exchange, left represents the most illiquid early markets, right the most liquid, late ones

Average CLV by Matched Volume:
vol_bin
<2000           31.017859
2000-5000       21.940813
5000-10000      15.790530
10000-100000    19.375408
>100000         19.256667

P&L and ROI by Matched Volume:
                 P&L       ROI    Bets
vol_bin                               
<2000         358.45  0.289540  1238.0
2000-5000     345.96  0.132653  2608.0
5000-10000    314.86  0.177486  1774.0
10000-100000  150.99  0.063844  2365.0
>100000       190.88  0.237413   804.0

Projected EV [the estimated EV that’s included in the notifications]

A great proxy for true EV/CLV is necessary for improving volume-weighted ROI [or real world ROI/ROI that takes your stakes on different selections into account]. It’s clear that the projected EV provides a great representation of the actual value of the bet.

Average CLV by Projected EV:
proj_ev_bin
10-12.5      0.106646
12.5-17.5    0.146104
17.5-22.5    0.186099
>22.5        0.261004

P&L and ROI by Projected EV:
                P&L       ROI    Bets
proj_ev_bin                          
10-12.5      185.56  0.072969  2543.0
12.5-17.5    414.91  0.156040  2659.0
17.5-22.5    245.48  0.142142  1727.0
>22.5        515.19  0.276984  1860.0

Comments & conclusions

Several interesting conclusions can be drawn from this analysis.

Early/illiquid markets appear to be more beatable than late, liquid markets. From a theoretical point of view this makes perfect sense, if you’ve ever been handicapping races you know that early prices are terrible [on average] at predicting true probabilities compared to closing numbers [where informed, volume-weighted opinions have been given time to enter the markets]. However, what’s interesting in our case/with our strategy is that we’re not using a fundamental model to back our gambling, but rather the Betfair Exchange as our ‘source of truth’ for finding the selections worth wagering on. The Exchange is itself illiquid hours out which could suggest some caution against using what’s happening there to beat ‘less informed’ counterparties. An invalidated fear, it seems…
Projected EV [the number included in the notification] is a great proxy for the true value of the bet. Helpful for assisting with staking [stake more when the EV is high, less when it’s low].
Bet365 is the best bookie for this framework. Kambi & 888sport shares the runner-up spot and then we’ve got Unibet a little bit behind [~12-13 % CLV so should still be very good on a long-term basis].
Throughout the analysis we’ve assumed flat stakes, pure/clean win bets [no place etc.] & that you’ll be quick enough to get all bets on. These assumptions are primarily used for ease of analysis & in real world applications you’ll of course see variable stakes [more on lower odds, more on higher EV], E/W-action [betting each-way rather than ‘win only’ is highly lucrative in ~25-30 % of the cases, get in touch for more info] and software/bot/latency/you-not-being-online issues that prevent you from being able to place all recommended bets. We’ve however been running everything ourselves LIVE since the launch of the Racing Bot to make sure it actually works not only in theory but also in practice. The verdict? It sure does…

For access to lower-latency, higher volume [~2-300 bets/day] feeds, contact us at betstreams@bowtiedbettor.com. Note that this option/version comes at a [significantly] higher cost.

Next week we’re back with a similar analysis of the US Sports Channel.

Stay toon’d!

Disclaimer: None of this is to be deemed legal or financial advice of any kind. These are *opinions* written by an anonymous group of mathematicians who moved into betting.

BowTiedBettor