Yeah, from what I can see it only affects the strike rate by a very small amount.
The obvious thing missing from the article of course is the actual available price at the point of entry. He says something like 'it snaps back to near the pre match price'.
From my data, I think there would been four qualifiers today (including the ranking filter). For this tiny sample this is what the pre match and one set all prices were.
Tennis article
I'd first try to replicate his findings by following his method as closely as possible, just to see if he's up to snuff.
It is possible, but that it's not where I'm focused at the moment. I'm just focused on looking at specific scenarios.
If I can sanity check that and keep on working on those specific scenarios to look for any potential errors in the wiring, then I'll move it to be a lot broader.
I need to get myself some tennis data. I used to have loads but for some odd reason I binned itEuler wrote: ↑Fri May 22, 2026 2:13 pmIt is possible that it's not where I'm focused at the moment. I'm just focused on looking at specific scenarios.
If I can sanity check that and keep on working on those specific scenarios to look for any potential errors in the wiring, then I'll move it to be a lot broader.
Tennis abstract has lots of data if you haven't collected any but obviously as we're coming into peak season for tennis now, you should be able to accumulate a lot using Bet Angel.
The biggest problem that I'm having, that you will have as well, is that as you wire up and create the logic, you start to find all sorts of errors and misinterpretations that you have to error trap before you can become confident that it's outputting the correct results. That seems to be the hardest bit.
I've actually written a whole set of logic to spot errors so that I can focus on that more than the actual outputted data itself. Once I'm confident that I'm spotting all errors and the data is clean and the output is clean, then I'll be able to post up some results.
I must admit I'm half expecting it to say that the market's pretty efficient.
The biggest problem that I'm having, that you will have as well, is that as you wire up and create the logic, you start to find all sorts of errors and misinterpretations that you have to error trap before you can become confident that it's outputting the correct results. That seems to be the hardest bit.
I've actually written a whole set of logic to spot errors so that I can focus on that more than the actual outputted data itself. Once I'm confident that I'm spotting all errors and the data is clean and the output is clean, then I'll be able to post up some results.
I must admit I'm half expecting it to say that the market's pretty efficient.
-
federernadal4ever
- Posts: 8
- Joined: Sun May 24, 2026 8:52 am
Very interesting. I have a tennis database since 2002 so will test it.
Okay so I didn't get to post it! Been too busy on RG26.
Here's what I think the article is saying: -
I've just read through this article and, in simple terms, the strategy is based on the idea that the market overreacts when a favourite recovers after losing the first set.
The setup is:
1. A player starts as the pre match favourite.
2. They lose the first set.
3. They win the second set.
4. The match goes to a deciding third set.
5. The market assumes the favourite has regained control and shortens their price.
6. The strategy lays the favourite at the start of the third set.
The theory is that punters and traders place too much emphasis on what has just happened. The favourite wins the second set and suddenly the narrative becomes "normal service has resumed". The author argues that, historically, favourites in this situation lose more often than the market expects.
So, for example:
Player A starts at 1.40 before the match.
They lose the first set but win the second.
Going into the decider they may be trading around 1.50 or shorter.
The strategy says that price is too short, so you lay the favourite and back the underdog indirectly.
---
What's interesting is that this is really more of a behavioural bias strategy than a tennis strategy. It relies on the idea that markets overreact to momentum and recent events.
The article itself is actually more interesting than the strategy. The author spends a lot of time discussing things such as avoiding overfitting, using out of sample testing, bootstrap testing and looking for logical reasons why an edge might exist rather than simply accepting a profitable backtest.
As always, a reported 35% ROI should be treated with caution. Tennis markets are generally very efficient, execution matters, commission matters, and any edge that becomes widely known can disappear over time.
That said, the underlying idea is plausible. Sports betting markets are full of recency bias, and people are often far too quick to assume a match has "turned around" after seeing one set of evidence.
Here's what I think the article is saying: -
I've just read through this article and, in simple terms, the strategy is based on the idea that the market overreacts when a favourite recovers after losing the first set.
The setup is:
1. A player starts as the pre match favourite.
2. They lose the first set.
3. They win the second set.
4. The match goes to a deciding third set.
5. The market assumes the favourite has regained control and shortens their price.
6. The strategy lays the favourite at the start of the third set.
The theory is that punters and traders place too much emphasis on what has just happened. The favourite wins the second set and suddenly the narrative becomes "normal service has resumed". The author argues that, historically, favourites in this situation lose more often than the market expects.
So, for example:
Player A starts at 1.40 before the match.
They lose the first set but win the second.
Going into the decider they may be trading around 1.50 or shorter.
The strategy says that price is too short, so you lay the favourite and back the underdog indirectly.
---
What's interesting is that this is really more of a behavioural bias strategy than a tennis strategy. It relies on the idea that markets overreact to momentum and recent events.
The article itself is actually more interesting than the strategy. The author spends a lot of time discussing things such as avoiding overfitting, using out of sample testing, bootstrap testing and looking for logical reasons why an edge might exist rather than simply accepting a profitable backtest.
As always, a reported 35% ROI should be treated with caution. Tennis markets are generally very efficient, execution matters, commission matters, and any edge that becomes widely known can disappear over time.
That said, the underlying idea is plausible. Sports betting markets are full of recency bias, and people are often far too quick to assume a match has "turned around" after seeing one set of evidence.
I ran the exact strategy on all of the markets at RAG26 so far and this is what I found.
No specific odds requirement, i.e. favourite is lower than 2.00 returns +£126 to a £100 fixed stake. Return is higher if you use my recommended liability staking.
No specific odds requirement, i.e. favourite is lower than 2.00 returns +£126 to a £100 fixed stake. Return is higher if you use my recommended liability staking.
Interesting you should post this today Peter.
I've been away the last week but seeing as I have no tennis history data with BF pricing, I've had my recorder code saving the BF streams for each match over the last 7 days. I have some metadata on players and can do some filtering by ATP\WTA, surface and event if needed, but today I created some quick code to test this out with no filter - given it's only a small sample
One issue is that I have no score data linked so have had to infer S1\S2 end and the start of S3 from the price moves.
So far it does look a little promising, but given the number of qualifying matches I have it could be a random result
One thing the code does throw up currently is that a range on the fave start price might be useful. i.e > x < y . A very strong fave tends to take the 3rd the majority of the time, as expected . And then another max price on the lay entry at the start of the 3rd
I'll keep recording the streams over the next few weeks and see if the results hold up, and if the thresholds change. And if rank etc filtering is worth pursuing
So far I've had 38 qualifiers with a return of £4.94 (£1 flat stake) at 28.7% ROI, 2% comm.
n=38
pnl=+4.94
roi_liab=0.287
I've been away the last week but seeing as I have no tennis history data with BF pricing, I've had my recorder code saving the BF streams for each match over the last 7 days. I have some metadata on players and can do some filtering by ATP\WTA, surface and event if needed, but today I created some quick code to test this out with no filter - given it's only a small sample
One issue is that I have no score data linked so have had to infer S1\S2 end and the start of S3 from the price moves.
So far it does look a little promising, but given the number of qualifying matches I have it could be a random result
One thing the code does throw up currently is that a range on the fave start price might be useful. i.e > x < y . A very strong fave tends to take the 3rd the majority of the time, as expected . And then another max price on the lay entry at the start of the 3rd
I'll keep recording the streams over the next few weeks and see if the results hold up, and if the thresholds change. And if rank etc filtering is worth pursuing
So far I've had 38 qualifiers with a return of £4.94 (£1 flat stake) at 28.7% ROI, 2% comm.
n=38
pnl=+4.94
roi_liab=0.287