Data Fluctuation on sample

[jimibt]

Morning all,

Not quite sure how to position this question but will take a stab. I've been devising an approach which i've been plugging small stakes into for the past 93 days (basically, the summer season). This has been pretty stable and the betfair stats have married up 100% to the bets that I've placed. So far so good.

However, despite it all being pretty rosy, i thought i'd chart out the historical data to see how it all looked. everything looks great and confidence still high. The only slight concern tho on graphing the data was that I noticed a rather more erratic set of behaviours between days 43 and 78. Basically, this region shows a wider deviation from the trendline and although the trend wasn't broken, I wondered if there was some adavnced analysis on Excel that would be able to identify whether the trend was about to drop?. My thinking is that round about day 52 (had i been charting at the time), i may have considered the trend to be breaking down. However, not only did it pull out of that nose-dive, it also by day 64 took on an equally large positive trend break (which returned to mean on day 80).

In short, had i not been using small stakes and had i actually charted this at the time, i may have been tempted to re-examine the approach and thus would have missed out on what appears to be a self resolving trend. Graph below - thoughts welcome:

[LinusP]

Where is the nose-dive?

[jimibt]

Where is the nose-dive?

Hi liam - nose-dive is potentially a bit of hyperbole. :d

i guess the question is centered more around the data being less uniform and less *on trend* across that 43-78 day window. I just wondered really if that type of flucuation could be measured in a less linear fashion, so as to not (at the time) to appear to be deviating significantly from the trend. my rationale here is based on this being a live strategy and therefore looking to confirm that the small blip (which was more pronounced at the time) is indeed not hugely out of alignment with the general trend.

call me a woos - just looking to the future on this for when a similar trend deviation occurs (which it will no doubt).

hope this makes sense.

[sa7med]

What sport is it running on? Perhaps nose dive is due to liquidity shifting to other more lucrative sporting events? I usually find my bot performing worse in such situations. What month did the dive take place? For me Aug want great due to, I'm guessing, punters on holiday. Could be a seasonal thing?Though this is my first year doing this and have in no way quantified this.

I also dont see a nose dive on your curve. Looking pretty good to me. I would say that if it's in profit after 90 days dont let it bog you down too much and perhaps focus on finding other edges

[jimibt]

What sport is it running on? Perhaps nose dive is due to liquidity shifting to other more lucrative sporting events? I usually find my bot performing worse in such situations. What month did the dive take place? For me Aug want great due to, I'm guessing, punters on holiday. Could be a seasonal thing?Though this is my first year doing this and have in no way quantified this.

I also dont see a nose dive on your curve. Looking pretty good to me. I would say that if it's in profit after 90 days dont let it bog you down too much and perhaps focus on finding other edges

it was indeed August (UK/IRE horse racing) - I actually wondered if it was due to the courses being overly hard due to the lack of rain in August and therefore affecting my prior analysis which forms the basis of this strategy. OPne thing I don't yet account for is the going, so may well be something i need to plug in.

as you say, maybe i'm bogging myself down with minutiae and shouldlook to further levelrage what is there, e=rather than focus on blips that only i seem to be focussing on

[LinusP]

Where is the nose-dive?

Hi liam - nose-dive is potentially a bit of hyperbole. :d

i guess the question is centered more around the data being less uniform and less *on trend* across that 43-78 day window. I just wondered really if that type of flucuation could be measured in a less linear fashion, so as to not (at the time) to appear to be deviating significantly from the trend. my rationale here is based on this being a live strategy and therefore looking to confirm that the small blip (which was more pronounced at the time) is indeed not hugely out of alignment with the general trend.

call me a woos - just looking to the future on this for when a similar trend deviation occurs (which it will no doubt).

hope this makes sense.

The strategy looks good, hard to tell but sample looks small, how many bets/markets?

For reference, this is a nose dive (occurred during the same period Aug/no rain):

2018_au789.png

[jimibt]

Where is the nose-dive?

Hi liam - nose-dive is potentially a bit of hyperbole. :d

i guess the question is centered more around the data being less uniform and less *on trend* across that 43-78 day window. I just wondered really if that type of flucuation could be measured in a less linear fashion, so as to not (at the time) to appear to be deviating significantly from the trend. my rationale here is based on this being a live strategy and therefore looking to confirm that the small blip (which was more pronounced at the time) is indeed not hugely out of alignment with the general trend.

call me a woos - just looking to the future on this for when a similar trend deviation occurs (which it will no doubt).

hope this makes sense.

The strategy looks good, hard to tell but sample looks small, how many bets/markets?

For reference, this is a nose dive (occurred during the same period Aug/no rain):

2018_au789.png

sample was for period 1st july to the present and totals 568 bets. Covers 70% of the UK/IRE courses. I tend to concur with the August lack of rain and it's one of those lightbulb moments both thinking about that and seeing you also mention it!!

[edit] - btw - my chart plots daily totals, rather than individual trades ... the daily average is about 6-7 bets, tho some days, none at all.

[Wolf1877]

I think you would have to start looking into probability theory and statistical confidence limits to see if your trading system had ceased to become effective.

For example if you made a weighted coin that had an exact 60 percent probability of landing heads (hence a 40 percent probability of tails) and each day you tossed the coin 10 times winning a pound every time it was heads and losing a pound when it was tails. If you did that for a 100 days then I'd expect you to be £200 up or thereabouts with your charted trendline going up £2 a day.

The exact figures on any day wouldnt always be £2 and the final sum is unlikely to be exactly £200 though. In fact the chances of you being £2 up on any individual day would only be approx 25%. The probability of you losing money on any one day would be approx 16.6% chance. The probability of you losing money on two successive days is approx 2.8% and the probability over any 3 successive days is just under 0.5% or 200-1 and there are So over 100 days you it is not massively unlikely that you might lose money 3 days in a row just through random fluctuations. If you lost money 5 days in a row though the probability of that happening through random chance is about 8000-1 so you might start to think that someone had tampered with your coin and after 6 or 7 consectutive days of losses then you'd start to get increasingly certain that something had changed. Obviously trading is a lot trickier than tossing a coin with a fixed bias but if you calulate an observed probability of you losing money on any individual race are x% then with some maths you can use probability theory to measure how unlikely any losing streaks that you chance upon actually are.

[jimibt]

wolf -for a minute there i panicked and thought you'd hacked my machine!! max losses are 3 days in a row, max wins are 5 in a row. normal pattern is more like 3 wins followed by a losing day...

just checking my avast AGAIN

[Wolf1877]

wolf -for a minute there i panicked and thought you'd hacked my machine!! max losses are 3 days in a row, max wins are 5 in a row. normal pattern is more like 3 wins followed by a losing day...

just checking my avast AGAIN

Well I did have a bit of a subconcious clue from the chart you uploaded. Maybe you should inspect your motherboard just to check if the chinese have added any spy chips though!

[foxwood]

I actually wondered if it was due to the courses being overly hard due to the lack of rain in August

That was first thought on seeing the graph in the OP and laughed when saw that post afterwards.

Have a look at splitting out graphs for Flat / NH / AWT - maybe that will help diagnose if is going related.

Graph looks very tight to trend to me - certainly one to run with a fair bit harder than you seem to be if the bank will take it.

Don't be scared of the dips - consider longest losing PL and depth ie from first loss to recovering back to same point - bank has to be able to cover at least that to get back on trend at any time - that is of course if its going to go back up

[jimibt]

I actually wondered if it was due to the courses being overly hard due to the lack of rain in August

That was first thought on seeing the graph in the OP and laughed when saw that post afterwards.

Have a look at splitting out graphs for Flat / NH / AWT - maybe that will help diagnose if is going related.

Graph looks very tight to trend to me - certainly one to run with a fair bit harder than you seem to be if the bank will take it.

Don't be scared of the dips - consider longest losing PL and depth ie from first loss to recovering back to same point - bank has to be able to cover at least that to get back on trend at any time - that is of course if its going to go back up

well, today was one of the 3 in a row losing days -lol. but this is obviously a great test to see if the averages are maintained. and yes, i agree that i should have ramped it up around the 24th day, but happy to spin it out on min stakes and compare to actuals, just to be certain that criteria is sound and selections are consistent with BF historical selections. next phase is the slow ramping up and will bear in mind your points regards allowing for the dip AND recovery, as opposed to just spinning out of the dip!! very good point.

thanks for inputs from all, will keep you posted on progress and feel free to ping further observations... as and when!!