No newly-created, or newly-purchased, selection system should ever be used in earnest until it has been fully tested and has proven itself to be profitable.

Testing provides a degree of reassurance that the system will be profitable in the future.

During the testing period, the selections generated by the system should be used to place ‘theoretical’, rather than actual, bets.

The following information should be recorded, during the testing period, for future analysis:

• Selection • Stake • Odds at the time that the bet would have been placed • Result • Profit/loss

The results should be periodically analysed and the strike rate and the profitability of the system determined.

If the system is to be used to back selections to win, the average odds of the winning selections should be calculated.

If the system is to be used to lay selections to lose, the average odds of the losing selections should be calculated.

At some point during the testing period, the decision as to when to begin using the system to place actual bets will need to be taken.

With regard to this, there are three options:

• At a random point in time • At a pre-determined point in time • After a pre-determined number of bets

All of the above three options suffer from the same issue in that the system may not yet be in a position to be used to place live bets since too little testing may have been performed.

All of the above options are, essentially, random and therefore not very scientific.

A far better option, in my opinion, is to take a more scientific approach to this issue.

Fortunately, this isn’t difficult.

Here is an example:

Let us suppose that a new system has been used to place 1,000 theoretical bets and the strike rate was found to be 79.85%.

The implication is that the true, long-term, strike rate of the system is 79.85%.

However, this may not be the case and the measured strike rate is only an estimate of the system’s true, long-term strike rate.

Whenever a strike rate is measured, only a relatively small ‘sample’ of all possiblepast results are used.

And there is the issue.

When we attempt to determine the future performance of a system in this way, the following assumptions are made:

• That the sample is reasonably representative of the system’s past performance. • That the system will perform in a similar way in the future to that in the past.

If the above assumptions hold true, and if the past strike rate of the system was 79.85%, then it is likely that the future strike rate of the system will be in the region of 79.85%. At least, that’s the theory.

Provided that the sample of 1,000 results is reasonably representative of the past results, then, the estimate of 79.85% is likely to be fairly accurate.

If the sample was un-representative of the past results, then the estimated strike rate will be inaccurate.

Now, because only a relatively small sample of all of the possible past results is used to estimate the strike rate of the system, there must be some doubt as to its accuracy.

If the strike rate is inaccurate and it is know how inaccurate it is likely to be, then the inaccuracy in the estimate can be corrected.

Fortunately, a method exists for estimating just how inaccurate an estimate is likely to be. All that is needed to calculate the inaccuracy is the sample size.

The mathematical terminology for the inaccuracy in the estimate is the ‘Margin Of Error’ or MoE.

The MoE is given by this simple formula:

MoE = 100/(square root of the sample size) In our case, the sample size is 1,000. The square root of 1,000 is 31.62. Therefore, MoE = 100/31.62 MOE = 3.16%

Therefore, the strike rate of the system, using a sample size of 1,000 results, is accurate to within 3.16% of its true, long-term, value.

The true long-term strike rate of the system is therefore likely to be as low as: 79.85 - 3.16 = 76.69% and is likely to be as high as: 79.85 + 3.16 = 83.01%.

So, we now know that the true, long-term strike rate of the system is likely to besomewhere between 76.69% and 83.01%.

Now, let’s assume that testing continues until there are 2,000 results and that the strike rate of the system is still found to be 79.85%.

How inaccurate is the strike rate of the system likely to be now?

The square root of 2,000 is 44.72. Therefore, MoE = 100/44.72. MOE = 2.24%

Therefore, the strike rate of the system, a sample size of 2,000 results, is accurate to within 2.24% of its true, long-term, value.

The true long-term strike rate of the system is likely to be as low as: 79.85 - 2.24 = 77.61% and is likely to be as high as: 79.85 + 2.24 = 82.09%.

What we see here is that the greater the sample size, the lower the MoE is and the more accurate the estimate becomes.

Another way of expressing this is that the greater the size of the sample, the more faith that we can be place in the estimate.

It is worth noting, however, that a doubling of the sample size does not bring about a halving of the Margin of Error.

If we use the above examples, although the sample size is increased from 1,000 to 2,000, the Margin of Error only decreased from 3.16 to 2.24.

So, even though the sample size was increased by 100%, the Margin of Error only decreased by 100% x (3.16 -2.24)/3.16 = 100% x 0.92/3.16 = 100% x 0.2911 = 29.11%.

In other words, an increase in the sample size does not give rise to a corresponding increase in the accuracy of estimates.

This is an example of the Law of Diminishing Returns.

What this shows is that the more testing that is performed, the more accurate the estimates are likely to be.

However, there comes a point where further testing does not appreciably improve the accuracy of the estimates.

OK, so we now know what the MoE is and we also know how to calculate it. But, how does the MoE assist us?

Let’s look at these three examples:

Example 1.

Let’s say that we have tested a new laying system and placed ‘theoretical’ bets on 1,000 selections. At the end of the testing period, the strike rate of the system was found to be 83.55% and the average betting exchange odds of the losing bets was found to be 5.0.

Firstly, let’s calculate the MoE: MoE = 100/(square root of the sample size)

The square root of 1,000 is 31.62.

Therefore, MoE = 100/31.62, MOE = 3.16%

Therefore, the strike rate of the system, measured over 1,000 results, is accurate to within 3.16% of its true, long-term value.

Therefore, the true long-term strike rate of the system is likely to be as low as: 83.55 - 3.16 = 80.39% and is likely to be as high as: 83.55 + 3.16 = 86.71%.

So, we now know that the true, long-term strike rate of the system is likely to be somewhere between 80.39% and 86.71%.

Secondly, let’s now calculate what the strike rate of our system needs to be given that the average odds of our losing bets is 5.0 (betting exchange odds):

Betting exchange odds of 5.0 equals fractional odds of 4/1. Fractional odds of 4/1 means that there are 4 chances of the selection losing and 1 chance of the selection winning.

The total chances are therefore equal to 4 + 1 = 5.

When laying a horse, in order to win the bet, the selection must lose.

There are 4 chances of it losing out of 5 in total. 4/5 = 0.8. To convert 0.8 to a percentage, we simply multiply by 100.

Therefore, 0.8 x 100% = 80%.

Therefore, if our laying system is to break even, the strike rate must be equal to 80%.

Otherwise, the laying system will lose money long term.

However, betting exchanges charge 5% commission on all wining bets.

Therefore, we must increase the required 80% strike rate by 5%.

To do this, we simply multiply the required strike rate by 1.05. So, 80 x 1.05 = 84.

Therefore, the strike rate of the laying system needs to be greater than 84% if a long term profit is to be made.

Now, we know that the actual long-term strike rate of the system is likely to be somewhere between 80.39% and 86.71%.

If the actual long-term strike rate is 80.39% (the lowest likely value), the system will make a loss since it is less than 84% (the break-even strike rate).

Therefore, if the system is used to place real bets, it could lose money, long-term.

At this stage, there are three options:

• Write the system off • Use the system to place actual bets. • Perform more testing.

With regard to the first option, if the strike rate of the system is as high as 86.71%, which is not beyond the realms of possibility, then we would be laying to waste a potentially profitable system.

At this stage, this option is not to be recommended.

With regard to the second option, we would be taking too much of a risk.

Yes, if the strike rate of the system is as high as 86.71%, then a profit will be made since the strike rate is above the 84% required to make a profit.

However, the system’s strike rate could equally be as low as 80.39% and a loss will be made since it is less than the break-even value of 84%.

This option is therefore not to be recommended either since it carries a certain amount of risk.

With regard to the third option, if we perform more testing, we will reduce the MoE and arrive at a strike rate which has less latitude and one in which we can have more faith that the measured strike rate is accurate.

Of the three options, we would recommend that the system is tested further.

Now, let’s change things slightly and look at this example.

Example 2.

Let’s say that we have tested a new laying system and placed ‘theoretical’ bets on 1,000 selections.

At the end of the testing period, the strike rate of the system was found to be 78.33% and the average betting exchange odds of the losing bets was found to be 5.0.

Firstly, let’s calculate the MoE: MoE = 100/(square root of the sample size). The square root of 1,000 is 31.62. Therefore, MoE = 100/31.62, MOE = 3.16%

Therefore, the strike rate of the system, measured over 1,000 results, is accurate to within 3.16% of its true, long-term, value.

The true long-term strike rate of the system is likely to be as low as: 78.33 - 3.16 = 75.17% and is likely to be as high as: 78.33 + 3.16 = 81.49%.

So, we now know that the true, long-term strike rate of the system is likely to be somewhere between 75.17% and 81.49%.

Secondly, let’s now calculate what the strike rate of the system needs to be given that the average odds of the losing bets is 5.0 (betting exchange odds):

Betting exchange odds of 5.0 equals fractional odds of 4/1.

Fractional odds of 4/1 means that there are 4 chances of the selection losing and 1 chance of selection winning.

The total chances are therefore equal to 4 + 1 = 5.

When laying a horse, in order to win the bet, the selection must lose.

There are 4 chances of it losing out of 5 in total. 4/5 = 0.8.

To convert 0.8 to a percentage, we simply multiply by 100.

Therefore, 0.8 x 100% = 80%. Therefore, if the laying system is to break even, the strike rate must be equal to 80%.

Otherwise, the laying system will lose money long term.

However, betting exchanges charge 5% commission on all wining bets.

Therefore, the required 80% strike rate must be increased by 5%.

To do this, we simply multiply the required strike rate by 1.05. So, 80 x 1.05 = 84.

Therefore, the strike rate of the laying system needs to be at least 84% if a profit is to be made.

Now, we know that the actual long-term strike rate of the system is likely to be somewhere between 75.17% and 81.49%.

If the actual long-term strike rate is 75.17% (the lowest likely value), the system will make a loss since the system’s strike rate is below the minimum 84% required to make a profit.

Even if the actual long-term strike rate is 81.49% (the highest likely value), the system will still make a loss since the system’s strike rate is still below the minimum 84% required to make a profit.

Therefore, if the system is used to place real bets, it will most probably lose money, long-term.

At this stage, there are three options:

• Write the system off. • Begin to use the system to place actual bets. • Perform more testing.

With regard to the first option, even if the strike rate of the system is as high as 81.49%, then it is highly likely that it will lose money since this is less than the break- even value of 84%.

At this stage, this option is probably the best one and is, therefore, to be recommended.

With regard to the second option, we would be taking too much of a risk.

Even if the strike rate of the system is as high as 81.49%, then it is highly likely that it will lose money long term because it is less than the break-even value of 84%.

This option is therefore not to be recommended since it carries too much risk.

With regard to the third option, if more testing is performed, the MoE will be reduced and we will arrive at a strike rate which has less latitude and one which we can have more faith in.

However, even if the strike rate of the system is as high as 81.49%, then it is highly likely that it will lose money since it is less than the break-even value of 84%.

This option is therefore not to be recommended since it carries too much risk.

Of the three options, we would recommend that the system is written off and not used any more.

Again, let’s change things slightly and look at this example.

Example 3.

Let’s say that we have tested a new laying system and placed ‘theoretical’ bets on 1,000 selections.

At the end of the testing period, the strike rate of the system was found to be 87.95% and the average betting exchange odds of the losing bets was found to be 5.0.

Firstly, let’s calculate the MoE: MoE = 100/(square root of the sample size). The square root of 1,000 is 31.62. Therefore, MoE = 100/31.62, MOE = 3.16%

Therefore, the strike rate of the system, measured over 1,000 results, is accurate to within 3.16% of its true, long-term value.

The true long-term strike rate of the system is likely to be as low as 87.95 - 3.16 = 84.79% and is likely to be as high as 87.95 + 3.16 = 91.11%.

So, we now know that the true, long-term strike rate of the system is likely to be somewhere between 84.79% and 91.11%.

Secondly, let’s now calculate what the strike rate of the system needs to be given that the average odds of the losing bets are 5.0 (betting exchange odds):

Betting exchange odds of 5.0 equals fractional odds of 4/1. Fractional odds of 4/1 means that there are 4 chances of the selection losing and 1 chance of selection winning.

The total chances are therefore equal to 4 + 1 = 5.

When laying a horse, in order to win the bet, the selection must lose.

There are 4 chances of it losing out of 5 in total. 4/5 = 0.8.

To convert 0.8 to a percentage, we simply multiply by 100.

Therefore, 0.8 x 100% = 80%.

Therefore, if the laying system is to break even, the strike rate must be equal to 80%. Otherwise, the laying system will lose money.

However, betting exchanges charge 5% commission on all winning bets.

Therefore, 80% strike rate must be increased by 5%.

To do this, we simply multiply the required strike rate by 1.05. So, 80 x 1.05 = 84.

Therefore, the strike rate of the laying system needs to be at least 84% if a profit is to be made.

Now, we know that the actual long-term strike rate of the system is likely to be somewhere between 84.79% and 91.11%.

If the actual long-term strike rate is 84.79% (the lowest likely value), the system will make a profit since it is greater than 4% (the break-even value).

Therefore, if we use the system to place real bets, it is likely that money will be won, long-term.

At this stage, we have three options:

• Write the system off. • Begin to use the system to place actual bets. • Perform more testing.

With regard to the first option, even if the strike rate of the system is as low as 84.79%, which is not beyond the realms of possibility, then we would be laying to waste a potentially profitable system since the strike rate is above the 84% required to make a long-term profit.

At this stage, this option is definitely not to be recommended.

With regard to the second option, we would not be taking too much of a risk if we were to put the system live.

Even if the system’s strike rate were as low as 84.79%, the system would still be profitable since it is greater than the break-even strike rate of 84%.

This option is therefore recommended since it only carries minimal risk.

With regard to the third option, we could perform more testing and the MoE will be reduced and a strike rate with less latitude, and one which we can have more faith in, will be arrived at.

However, if we do this, we would have to forego the potential profits that the system would make otherwise make.

This is perhaps the safest option, but, at some point, all systems have to pay their way, reward us for our efforts and go live.

Of the three options, we would recommend that the system is put live.

Now, let’s summarise the above steps involved in the determination of when, and if, a system should go live, just in case there is any confusion.

• Test the system using theoretical, rather than actual, bets. • Calculate the strike rate. • Calculate the MoE. • Subtract the MoE from the strike rate to give the lowest probably strike rate. • Calculate the average odds of the losing bets and convert it to a strike rate. • If the lowest probable strike rate is greater than the strike rate required to make a profit, then the system can go live. Otherwise, consider more testing or writing the system off.

There is one other proviso.

It is recommended that no new system should ever be used to place actual bets unless it has been tested on at least 1,000 theoretical bets.

When the system is eventually used to place actual bets, as opposed to theoretical ones, we would recommend that minimum stakes are used initially and that the stakes are increased slowly.

This will ensure that any mistakes that are made will have minimum impact.

At this stage, it would be far better to have patience rather than an empty bank account!

So, from this article, we have learned about the Margin of Error associated with a system’s strike rate, how to calculate it and how it may be used to determine when a system can be safely used  to place actual, as opposed to theoretical, bets.