Following on from the Premier League Fund, I have adapted the algorithm to the Championship.  As well as adding profits, the addition of a second Fund should also reduce the overall volatility of total invested funds.  This is Markowitz portfolio theory working its statistical magic.  Volatility is reduced since the performance of the Premier League and Championship models are completely uncorrelated.

A word here on yields and returns.  Whilst the additional profit from a second Fund does not increase yields, per se, it can increase percentage returns of total cash invested.  Yields are calculated as net profit from stakes placed divided by the total of those stakes.  Fund returns however are calculated as total profits divided by cash held (the Fund balance) regardless of whether this cash is used to place bets; i.e. put at risk.  The cash in the Fund spends most of its time unused, earning zero interest.  Therefore the more bets we can be place, the better use we make of our cash and the higher the overall returns.  The issue comes of course that a lot of games are played at the same time.  The Trading Fund will however take advantage of games which are at different times.

Additionally, it may be possible to remove cash from betting accounts to earn bank interest when it is not being used.  On the downside this adds to administration but with interest rates on instant access accounts currently at attractive rates it should be worthwhile.

Performance

Before starting work on the Championship algorithm I was of the opinion it would offer less profit as 6 teams changed season to season; 3 relegated into and 3 promoted out of the league.  However, based on backtested results, there do appear to be sizeable gains available¹.  Indeed, the returns are higher than for the Premier League Fund.  One reason could be that, compared to the Premier League, the Championship has a smaller following.  And so, if betting volumes and research are lower, the odds on offer in exchange markets may be less accurate.

In the table below are the model results using data backtested over the 5 seasons 2003/4-2007/8 with varying stake sizes.  Note we only have 5 seasons of data compared to the Premier League Fund’s 8 seasons.  This makes the results less reliable.  As for the Premier League data, the Fund starts at £100 and the final 2 seasons use Betfair data with the earlier seasons using a price exchange construct from traditional bookmakers’ odds.

Max. Stake size –>	1%	5%	10%
Positive/negative seasons	5/0	5/0	4/1
Winning bet % per match	48.12%	48.12%	48.12%
Final Fund value after 5% commission	£270	£6,237	£55,159
Mean return per season*	23%	159%	516%
Max. loss of Fund in any one season	+8%	+13%	-32%
Max. gain of Fund in any one season	43%	426%	1877%
Maximum exposure any one bet**	1%	5%	10%
Average number of matches before new high***	62	77	102
R-squared****	0.9741	0.9674	0.9526

*Note season returns do not follow the normal distribution due to % stake betting.  As for a normal distibution, where 95% of expected returns lie within -/+2 standard deviations of the mean, this is not the case for the model’s returns.

**Max. potential loss from any one bet as % of Fund.  This may be below the max. stake size due to value betting; i.e. each bet is reduced based on the amount of value the price is calculated as offering.

***A measure of how long the Fund ’stays under water’.  A figure of 40 would suggest the Fund is expected to reach a new high every month (10 matches/week).

****The r-squared measures the tightness of fit between the Fund balance and an exponential line of best fit over time.  A figure of 1 is a perfect fit, 0 no fit at all. As all gains are reinvested and the model uses % stakes we would expect the returns to be exponential.  The r-squared is a measure of how close the returns are to this exponential line of best fit. 

The vertical axis of the above graph is a log scale hence the straight line for the line of best fit, indicating exponential returns for the Fund.

Observations

What is apparent from the results is that the returns are higher compared with the Premier League Fund.  Though a lot of the gains come from one season in particular, a gain of 1877% at the 10% stake level.  Taking out this highest return and the lowest return seasons leaves an average return per season of 245% for the remaining 3 seasons.

The percentage of bets won is around 48% which means there is high volatility in the Fund balance.  Over time of course with value betting (assuming the algorithm is better than the market) we would expect to earn profits.  A low percentage of bets won coupled with large value available in the market odds helps to explain why we have 1 negative season and 1 extremely positive season.  What is reassuring is the average amount of time the Fund spends under water.  This is lower than the Premier League Fund suggesting that losing streaks (which we must expect in betting) are quickly outweighed by positive returns.

Conclusion

I will adopt a 5% staking plan for the Championship Fund as this season will be the first opportunity to place ‘live’ bets.

A final thought…..

Last season was the first time bets were placed live.  Whilst those bets did return considerable gains, most of the returns referenced on this site are from backtested results.  The financial markets have seen the dangers of relying on backtested results, not least in the recent repricing of credit risk.  Even Nobel Scientists have seen their sophisticated models fail by virtue of under-estimating correlation between markets. 

The issue for Funds in financial markets is two-fold.  Firstly, if you have identified mispricing and therefore an opportunity for profit you can be damn sure other participants have as well and therefore profits will be shared and diminish over time.  Secondly, backtesting by its nature looks at historic correlations.  Correlations between instruments vary over time and significantly once trades attempting to profit are put in place, something which George Soros is constantly trying to ram home.  In my opinion profit is earned in two ways; knowing more than everyone else in the market and putting in more work than everyone else in the market.  It’s analogous to exams, only in finanical markets, everyone sitting the exam after you gets to determine your grade.

And the relevance to this site? Well, i’m not concerned by correlation since I am looking at sports gambling where the bet is short in duration and the outcome is not influenced by the price of the selections. But the second issue does concern me.  If there are profits to be had from algorithm betting then am I not encouraging the dissemination of those profits by advertising such on this site.  Once again, only time will tell.

¹ Take a look at this post on Monte Carlo forecasting for consideration of the risk in using backtested results.