As the article on the importance of having a selection system demonstrated, if the past results of a selection system are stored, they can be analysed.

If the past results are capable of being analysed then the reasons for the past losses are capable of being determined.

If the reasons for the past losses can be determined, then any trends associated with the losses can be determined.

If the trends associated with losses can be determined, then filters can be created and applied to a selection system.

If filters can be applied to a system, then the possibility exists that future potential losing bets can be reduced in number.

If the number of future losing bets can be reduced, then the system’s strike rate will increase as will the system’s profitability.

Essentially, a filter is a rule which is applied to a selection system in order to eliminate selections that, based upon the analysis of past results, have a significant chance of providing us with a losing bet.

Filters can be based upon any variable associated with a horse.

For example, the jockey, trainer, race distance, course and going etc. can all be used as filters.

By way of an example, let us suppose that our system involves backing favourites to win in maiden races.

The analysis of the past results shows that the majority of the losing bets produced by the system were run at Sandown Park and Ascot.

The analysis also shows that a large proportion of losing bets related to races that were run on soft ground.

As a result of the analysis, the following three filters were identified:

• Track = Sandown Park • Track = Ascot • Ground = Soft.

These can be added to a system in such a way that if a selection is running in a race on soft ground or at Sandown Park or Ascot, it is eliminated and not backed to win.

As a result of implementing the three filters, the only favourites that will be identified by the selection system in future will be those running in maiden races at tracks other than Sandown Park and Ascot and that are run on ground other than soft.

The filters, once implemented, should reduce the number of losing bets, improve the strike rate, and hence the profitability, of the system.

Some of the most popular filters in use are odds based.

In some cases, a lower odds limit is specified. In some cases, an upper odds limit is specified.

In other cases, an odds range is specified.

The latter two are especially popular with laying systems.

Typically, a selection is only layed provided that the Starting Price is greater than, or equal to, evens and less than, or equal to, 4/1.

Another popular filter is to only lay a selection if the betting exchange odds are 5.0 or less.

In general, the average odds of the selections eliminated by non-odds-based filters are very similar to those of the remaining selections.

As such, non-odds-based filters, generally, have little to no effect on the average odds of the horses selected by the system.

This is not the case with odds-based filters which operate in an entirely different manner.

Odds-based filters do have an effect on the average odds of the remaining selections.

That’s the whole point of odds-based filters.

To illustrate the above, let us look at the following example:

Odds Based Filters

From the above table, we see that there are ten selections whose odds lie between evens and 10/1.

We can also see that the average odds of the ten selections are 6/1.

In order to mimic the effect of a non-odds-based filter, let’s remove selections 1, 3, 5, 7 and 9.

The table below shows the result of having removed the five selections.

Selections Removed

As can be seen from the above table, five selections remain. We can also see that the average odds of the remaining selections is 6.2/1.

This value is approximately equal to the average odds of the selections contained within the previous table i.e 6/1.

This shows that non-odds-based filters, in general, do not appreciably change the average odds of the selections generated by a system.

In order to mimic the effect of an odds-based filter, let’s remove, from the first table, all those selections whose odds are not greater than evens and greater than 4/1.

The following table shows the remaining selections:

Remaining Selections

As can be seen from the above table, three selections remain.

The other seven selections were removed by the odds filter.

We can also see that the average odds of the remaining selections are 3/1. This is approximately equal to half of the average odds of the un-filtered selections (6/1) contained in the first table.

This shows that odds-based filters will almost certainly change the average odds of the selections generated by a system.

As was stated earlier, filters are designed to reduce the number of future potential losing bets.

In the case of non-odds-based filters, the profitability of the selection system will probably be increased because the number of losing bets will probably be reduced.

This will improve the strike rate which, in turn, leads to an increase in the profitability of the system.

In the case of odds-based filters, the profitability of a system is rarely, if ever, improved.

Here is the explanation as to why this is:

Firstly, let's look at this issue from a backer’s point of view:

If horses are backed to win and a filter is applied to the system such that all of those horses whose odds are above a given limit are eliminated, those selections that are below the limit will remain.

As was stated in the article on betting exchanges, a horse’s strike rate is proportional to its odds.

Therefore, if those horses whose odds are below a given odds limit are selected, the strike rate of the selection system will increase because the strike rates of shorter priced horses are higher than those of the longer priced ones.

As a result, the frequency with which bets are won will increase.

However, due to the lower odds of the selections, less money will be won per winning bet than would have been won had the longer odds selections been backed to win.

In other words, by applying an odds-based filter, the win frequency will increase but the amounts won per winning bet will decrease.

Financially, therefore, nothing will be gained by applying an odds-based filter.

If horses are backed to win and a filter is applied to the system such that all of those horses whose odds are below a given odds limit are eliminated, those selections that are above the limit will remain.

As was stated in the previous article, a horse’s strike rate is proportional to its odds.

Therefore, if those horses whose odds are above a given odds limit are selected, the strike rate of the system will decrease because the strike rates of longer priced horses are lower than those of the shorter priced ones.

As a result, the frequency with which bets are won will decrease.

However, due to the higher odds of the selections, more money will be won per winning bet than would have been won had shorter odds selections been backed to win.

In other words, by applying an odds-based filter, the win frequency will decrease but the amounts won per winning bet will increase.

Financially, therefore, nothing will be gained by applying an odds-based filter.

Now, let's look at this issue from a layer’s point of view.

If horses are layed to lose and a filter is applied to the system such that all of those horses whose odds are above a given odds limit are eliminated, those selections that are below the limit will remain.

As was stated in the previous article, a horse’s strike rate is proportional to its odds.

Therefore, if those horses whose odds are below a given limit are selected, the strike rate of the system will decrease because the strike rates of shorter priced horses are higher than those of the longer priced ones.

As a result, the frequency with which bets are lost will increase.

However, due to the lower odds of the selections, less money will be lost per losing bet than would have been lost based filter, the loss frequency will increase but the amounts lost per losing bet will decrease.

Financially, therefore, nothing will be gained from applying an odds-based filter.

If horses are layed to lose and a filter is applied to the system such that all of those horses whose odds are below a given odds limit are eliminated, those selections that are above the limit will remain.

As was stated in the previous article, a horse’s strike rate is proportional to its odds.

Therefore, if those horses whose odds are above a given limit are selected, the strike rate of the system will increase because the strike rates of shorter priced horses are higher than those of the longer priced ones.

As a result, the frequency with which bets are lost will decrease.

However, due to the higher odds of the selections, more money will be lost per losing bet than we would have been lost had shorter odds selections been layed to lose.

In other words, by applying an odd-based filter, the loss frequency will decrease but the amounts lost per losing bet will increase.

Financially, therefore, nothing will be gained we will be gained from applying an odds-based filter.

From the above, it can be seen that applying odds-based filters to a selection system is of no financial benefit whatsoever since a change in the strike rate brings about a corresponding change in the amounts won and lost.

As the strike rate increases, the amount won per selection decreases.

As the strike rate decreases, the amount won per selection increases.

So, this begs the question ‘why do some backers and layers apply odds filters’?

There are several possible answers to this question:

Firstly, in my opinion, a number of those people who place bets on horses may not be as fully aware as they need to be of the above i.e. that although an increase/ decrease in the average odds of selections brings about a change in the strike rate of a system, it is offset by the change in the amounts won or lost per bet.

As a result, the implementation of an odds limit is mistakenly thought to bring about an increase in the profitability of a system.

Secondly, as the odds of a horse increases, the difference between its Starting Price (SP) and the odds on the betting exchanges grows exponentially.

The following table shows the relationship, in general, between the SP and the betting exchange odds.

Relationship Between SP and the Betting Exchange Odds

Although the SP has been used for some years now, betting exchanges are a relatively new concept.

As such, the SP has become the de facto standard. It is therefore assumed that the SP is accurate and the betting exchange odds are high.

In fact, this is not the case. As was stated in the bookmaker article, bookies reduce the odds of horses in order to create their profit margins (referred to as overround).

Therefore, in reality, it is the exchange odds that are the most accurate and not the SP (which is lower than it ought to be). Having said all this, the SP has been around for a long time now.

It is the standard that most people use and there’s no getting away from it.

Thirdly, and this particularly applies to those who lay horses to lose, the higher the odds, the greater is the liability on a bet.

If the odds become too high, then the risk is that the liability of the bet will exceed a sensible proportion of the betting bank.

This is discussed further in a future article.

In order to limit the liability of a bet, therefore, a limitation is placed on the odds of the horses identified by a system.

Before we move on to the next topic, there is just one other facet of this topic that we need to discuss.

The purpose of a filter is to reduce the number of future losing bets by identifying, and eliminating, those selections that will cause losses to occur.

When a filter is applied, the number of horses identified by the system will, by definition, decrease.

The greater the number of filters used, therefore, the greater will be the reduction in the number of selections.

It is therefore highly probable that, if too many filters are applied, all of the horses identified by the system will be eliminated and there will be none left to back or lay.

Filters should, therefore, be used sparingly.

In this article, we learned about filters: how they are created and why they are used.

We learned that, essentially, there are two types of filters: odds-based and non-odds- based.

We learned that non-odds-based filters improve the profitability of a system whilst odds-based filters generally don’t.

We also learned that filters should be used sparingly since, if too many filters are used, all of the horses identified by the system could be eliminated.