This one is amazing, it will save you a fortune that would otherwise be lost pursuing systems and staking plans that will not deliver long-term.

1. Use absolutely any selection methods and as many different ones as you like.

2. Paper trade your methods/selections for 1 year or 3,000 bets.

3. For each method record the price taken and Betfair SP of all selections.

4. Keep a running total of actual winners, expected winners, 'Act-Exp', 'Act-Exp less commission if positive', 'Act/Exp' & 'Chi-square'.

5. If A/E is greater than 1 at end of period, use the 'Act-Exp less commision if positive' total and Exp total to calculate POT (total of 'Act-Exp less commision if positive'/total of 'Exp').

6. If POT is greater than 2% at end of period, use the Act-Exp column to calculate a 'safe' number of units to divide your bank into.

7. Bet for real to return 1 unit of bank for a further year or 3,000 bets.

8. At the end of year 2 (or 6,000 bets), repeat step 5.

9. At the end of year 2 (or 6,000 bets) analyse your A/E by price band to give an indication of your edge in each band.

10. At the end of year 2 (or 6,000 bets) calculate growth in bank, profit and POT.

11. In year 3 or 6,000+ bets use 'conservative edge'/odds as your percentage of bank to stake. Conservative edge is based on analysis of edge by priceband.

12. Throughout the whole process monitor A/E to ensure it stays above 1. If it falls below 1 abandon the method.

13. Throughout the whole process monitor the Chi2 figure. If it stays above 10% after 3,000 bets abandon the method.

14. If bank growth during year 3 falls below year 2 adjust 'conservative edge' figures.

The post basically outlines the approach I found worked for me Michael.

'Expected' (e) = 1/decimal price and is also the stake that would return 1 point

Sum of Expected = expected winners (E) also the sum of proportional stakes to return 1 point

Actual = actual winner (a) = 1

Sum of actual = total winners (A)

As we all know, with a successful system/method the total winners divided by the total expected winners 'A/E' will be greater than 1. So monitor this. If it's less than one, any profit is due to the way your stakes have landed on the winner:loser sequence. You would have over-staked some winners and eventually the win:lose sequence will settle down to normal proportions and profits will vanish.

If there was no commission to pay then a stake of 1/price (e) would return 1 point (a) so profit on a winner would be 1 (a) minus expected (e) or 'a-e'. However we must reduce this by commission for a winner hence a - 'e less commission'. Losers obviously cost -1 x expected.

Total net profit would be the sum of winners A minus the sum of expected winners (or stakes) E less commission (A-'E less commission')

If we divide this total profit (A-E 'less commission') by total stakes (E) we get % profit on turnover POT.

If we divide this total profit (A-E less commission) by our starting bank we get % bank growth.

Chi2 is just a statistical measure of how likely it is you would have got he results by chance. It takes into account the number of bets and is not significant until it reaches under 5%. In my experience if under 10% and trending downward the method is worth persisting with.

Effectively the above is just saying,

In a successful system = A/E>1

So, there will always be a gross profit of A-E

but you would have to deduct commission from winning races as you go

Use Chi2 to monitor how sustainable A/E>1 might be

Monitor POT and note how important turnover is as POT gets smaller

If using betting as capital growth investment, the measure of success is ultimately bank growth. So in year 2 see how this is affected by the number of units of starting bank compared to the 1 unit you are staking to return.

Points 11 & 14 (year 3+) are just understanding Kelly in a practical way. Scale your stakes as a percentage of bank to the likely long-term proportions and more importantly potential sequences of winners and losers. This way you won't go bust and volatility can be controlled so growth is steady and greater than staking to return 1 point.

Excellent advise Andrew although I suspect many of us (me included!) do not have the patience to paper trade for a full year.

Also there is a risk that a system has a finite life (I think Nick Mordin suggested 3 years). By paper trading for a year you potentially lose a third of the benefits of a particular system.

The question is when do you have confidence that a system will be successful and when is it time to abandon it. The method you advocate is one way of answering this question.

Is the choice of 3,000 bets statistically significant? I think on other threads you have suggested 100 winners as being significant.

Now, using some sort of common sense, could you compromise by testing backwards for 18 months 2 yeears (I think we all know the perils of backfitting) - completing those calculations, then testing going forward for 12 months (or for one whole season of racing

)??

I tend to develop a system based on 5 years data and then test it on a full year's data. If it produces a profit on this test data then I would be more confident of using it live. Unfortunately even then I have tended to end up with systems that then do not make a profit live. I am still looking for the holy grail !

Hi Michael,

There's nothing significant in the 3,000 bets, it's just roughly the number I placed annually on my most successful method. The 100 winners is Michael Wilding's 'rule of thumb' for the minimum sample before beginning serious analysis.

I built my method on 10 years of past data and tested on a year or two of hold-out data. I then paper traded a year live and ran it and variations on it, for real for about eight years. It was and may still be profitable but was becoming really hard work as it involved compiling ratings and lookup tables and running automatic market monitoring and bet placement software. It was falling over quite regularly in the end (internet connection and my poor coding) so I used the bank to pay off the mortgage early. I had one of those interest only mortgages where the endowment was coming up hopelessly short of the capital so it made sense.

I ran other less successful methods for shorter periods but once I got the gist of how betting works* (probability, law of large numbers, edge, bankroll, volatility, staking etc) I realised it was inevitable they would all encounter a 'black swan' sooner or later. Double figure losing streaks do eventually happen with 50% strike rate systems. Perhaps that is what Mordin was alluding to.

I rarely bet these days and have no live 'systems' although I am trialing about a dozen approaches currently. It's a constant process of dropping failed ones and adding new. I reckon the 'one-year paper trading rule' has saved me more than I would have made during the short purple patches but I appreciate it's not for everyone.

* I still don't fully understand it and I'm still learning (I wish I'd paid more attention in Maths at school).

Thanks Andrew.

I must admit I've given up on form based systems, I became a bit disillusioned with the data issues with my handicap chase system and gave up in the end.

I've gone back to concentrating on the stock market as this is what I make my living from. Fortunately last year was a very good year with a 40% return on my investments. You still need balls of steel though on days like today where I am down over £10k. The black swan here is the Chinese flu outbreak.

The advantage of the stock market is that the average investor will make a profit. With horse racing the average punter will lose money.

I only bet now where bookmakers offers give me an edge. I am working on a system of backing odds on horses using the "wisdom of the crowds" approach where the odds quoted by one of the bookies appears to give an edge. Early days yet but in the last few days there have been 8 winners from 8 selections.