- This topic is empty.
-
Posts
-
Hi All,
It’s a good time to take stock.
The system has had a decent run for the last 30 days, see Profit/Loss chart (Simulation) (from 15th July to 15th Aug). We’re now fairly confident that we know why this has changed.
We’ve created an abridged version of our “Change log” (not the Full Monty, just a summary), which can be seen here: Change Log. This explains some of the work that’s been going on in the background, to help us understand why the system had not previously been living up to expectations.
There are 2 recent developments that have really helped:
- 13th July – We started using prediction models trained on the pre-race prices – not BSP
- 29th July – We built a Diagnostic system
Training models on pre-race prices, i.e. the prices at the same time our live system processes bets, seems to have had a significant impact. Arguably more important than that, was the development of our Diagnostics System. This has helped us narrow in on specific issues that have led to the Live system under-performing, when compared to our prediction models. 95% of the issues identified, have now been resolved, the remaining 5% we will have to live with (which is OK).
We’ve had to remove some predictors from our models – where we could not get a reliable feed. We’ve had to write new tools to help the system detect, and incorporate last minute jockey changes, and much more.
The key output from the diagnostic system is a profit loss graph for each model – as applied to both Simulation data, and live betting data. For the first time, these 2 graphs are now aligning – which means, the Live system should now start to follow our Simulation models. The chart that we’re using to track this is here: Live v. Simulation.
Current prediction models are estimating profits of around +240 points per month. We hope to achieve at least 80% of that. However, as we move into the Autumn/Winter, the number of races per day will reduce, which will of course reduce predicted profits.
In summary, now that we know the Live and Simulation systems are aligned, and whilst we continue to generate decent models, we expect to continue betting with real money for the foreseeable future.
All the best,
Nick
- The topic ‘Taking Stock’ is closed to new replies.