How useful is yield as a performance measure?

There are many variables to measure the success of a betting system.  These range from the simple ’strike rate’ showing the percentage of bets won to ‘yield’ which, by incorporating the potential loss on a bet, includes some measure of risk.  It goes without saying that the longer a system has been in place and the more bets it has covered the more reliable it is in terms of predicting future results.   However given the same number of bets, a system with a high yield is not necessarily better than a system with a lower yield.  Betting against Derby is a low yielding bet since the risk of loss is low.  However if such bets offer positive value (i.e. net profit over time) and we exclude them simply because they are low yield we are giving up potential profits.

It’s not just about risk (always calculated from historic figures) but also what might occur in the future

Using historic data I can generate a model which earns a very high return (on historic bets).  Breaking down the model however shows it achieves this, to a great extent, by betting against (laying) weak teams; e.g. Derby.  On the face of it, there appears to be value in laying weak teams.  Put another way, the odds on certain weak teams were overpriced by the market.  Extending the logic forward would suggest we bet big against these weak teams in future bets.  But the model only knows historic data.  It cannot incorporate future results - and therefore does not price the unknown.  It does not incorporate any event beyond odds of 16.  The longest odds on a team to win a match within the Premiership (as far as I know) was 16 for Man City beating Man U just last week on 10 Feb 08.  The model can now take this highly improbable and highly significant event into account.  It will now allow for such rare events.

It is analogous to pricing flood insurance based on assessing the flood damage over just a few years.  This period of time is too short to expect to include the high floods (and indeed low or no floods).  Worse still, it also ignores the possibility of floods getting worse year on year; e.g. from climate change.  Whilst I cannot see climate change affecting Premiership results perhaps the game is changing over time due to money or even playing games abroad!

The upshot of all this is simple but inconclusive.  We can only price based on what we know but we must be prepared for events outside the historic results.  We must expect the unexpected.  Inconclusive because by it’s nature we cannot accurately price the unexpected.

Life on the edge or dull but ‘predictable’

In response to the above I have developed 2 models and will track 2 funds, ‘Aggressive’ and ‘Balanced’.  The aggressive model uses only historic information and bets accordingly. It takes no account of the possibility of an event with odds wider than 16 (as priced by the market) occuring (currently).  It therefore takes greater risks and is more volatile.  It will likely earn higher profits in the short and medium run.  In the very long run it is expected to earn the same return as the balanced fund (be that positive or negative).  The second model is balanced to take account of the possibility of unknown events and has a higher yield.  It assumes events less likely than those which have already occurred are possible.  It does this by artificially limiting the calculated performance gap between the best and worst teams.