I am trying to build a system. I have results starting from 15 March this year to now. 533 selections.
return 79% for win and 83% place using Unitab prices.

Analyzing the results seems to throw up some odd things.
eg.
Race No.3 44 selections 16% S/R for a return of 39% at Unitab prices
Race No.2 42 selections 45% S/R for a return of 131% unitab prices.

Looking at these two races closer
R2. 50% of the winners are three year olds (they make up 28% of all the selections in R2)
R3. only 13% winners are three year olds but they make up 50% of all the selections

So the Question is do you (going by statistics) back all three year olds in race two and lay them in race three? or do you need more results?

If you are discriminating by age I would suggest that you would need data from each month of the year.
A 3yo in Sept is obviously not as mature as one at this time of the year.

Not so much as discriminating just an observation. The three year olds are returning 102% for metro and only 58% for non metro.(151 runners) but do agree with you about maturity.

Stats often show some very interesting observations.
It raises the question of how far into the future one can rely on them to sustain the statistical average .

This is where one has to have a cut off point as to volumn of losses that should be entertained before we cease betting until another result shows up.

Here are some other observations that I have found.

Race No.1 has the highest SR in RadioTAB top 2 than any other races No.

Any Hurdle race in Melb Races No.2 only has a 79% place SR on the market Fav,

Try & keep in mind that the stats can often change in the following year.

Just like the ever shifting sands of the dessert.

Cheers.

__________________
Cheers.

This has crossed my mind, you would think it should even out over time (how much time? how many races?) Not betting anything at the moment still collecting data.

As an exercise it would be interesting to see if an idea that works well over 4 previous years then have the data available to see if the idea worked say 10 years ago.
I feel not many systems would survive the test.

I found that a min of 4 years data is the min to work off, if checking to see if an idea will possible fly in the future.

If one is into researching ideas, it is hard to go past the brilliant GTX
It will cost money to use it, approx $100 a week for updates & day to day data.
Past results for each year can be purchased going back approx 9 years.

I found many ideas that worked well over say 3 years then fell in a hole on year 4.
Thats the weird thing in the fascinating world of system research.

"Why is it so?" one may ask,to coin an old popular saying from the much loved, professior Sumner Miller .

Cheers.

__________________
Cheers.