Can anyone, with some statistical and trading nous, suggest why the following consistent random procedure is not producing anything approaching a normal distribution?
£20 stake.
1. Random number generator Odds/Even number determines opening position Lay/Bet.
2. OFFER Lay/Bet to open. If OFFER taken and no move in position beyond one tick, for or against, within 10 secs then close position for small loss or 0 and discard the loss figure from final total! Re-submit OFFER in line with random selection.
3. After OFFER taken and movement greater than tick takes place then count 15sec.
4. Then attempt to close position by OFFERING at current price for 10sec. If odds move against within the 10sec then Take price to close position.
5. Repeat 1-4 until race countdown = 0
Average number of procedures (steps 1-4) = 6 per market. Market always Favourite. Begin time = £120K+ matched. End time= best fit to Start Time (countdown 0)
Individual market results:
£
-3.75
1.23
-1.62
-1.85
-2.89
-0.36
-1.76
0.49
-1.06
-1.55
-1.59
-5.16
-6.31
-2.20
2.97
-0.78
-3.65
-8.25
1.90
0.07
3.92
-2.35
-2.25
-0.70
-0.97
-0.28
-1.82
0
-1.85
-1.70
-3.02
0
-3.45
0.73
-0.85
-3.34
0
-1.54
-3.87
-2.66
-1.07
42 Markets 35 (83%) losses 7 (17%) wins
-£74.0 £11.31
Further non sense!
I haven't plotted your points but Initial thoughts without thinking too hard. I think firstly your sample size is too small for a normal distribution to necessarily show. Distribution is the frequency of occurrence at every profit/loss point. Also, because of adverse selection, when offering a price at random, your money will more likely be taken when the market is moving against you and therefore your curve's mean will be in the negative area. So even though you are generating your offers randomly, the fact that you are only counting the bets that are taken is injecting a bias. This is a typical feature of the spread. Additionally, there is the fact that different tick sizes offer different % profit/loss - so maybe it's better to be counting ticks rather than profit if you want to plot a distribution. Hope that helps.bastuncriochnatha wrote: ↑Sun Jan 26, 2020 1:05 amCan anyone, with some statistical and trading nous, suggest why the following consistent random procedure is not producing anything approaching a normal distribution?
£20 stake.
1. Random number generator Odds/Even number determines opening position Lay/Bet.
2. OFFER Lay/Bet to open. If OFFER taken and no move in position beyond one tick, for or against, within 10 secs then close position for small loss or 0 and discard the loss figure from final total! Re-submit OFFER in line with random selection.
3. After OFFER taken and movement greater than tick takes place then count 15sec.
4. Then attempt to close position by OFFERING at current price for 10sec. If odds move against within the 10sec then Take price to close position.
5. Repeat 1-4 until race countdown = 0
Average number of procedures (steps 1-4) = 6 per market. Market always Favourite. Begin time = £120K+ matched. End time= best fit to Start Time (countdown 0)
Individual market results:
£
-3.75
1.23
-1.62
-1.85
-2.89
-0.36
-1.76
0.49
-1.06
-1.55
-1.59
-5.16
-6.31
-2.20
2.97
-0.78
-3.65
-8.25
1.90
0.07
3.92
-2.35
-2.25
-0.70
-0.97
-0.28
-1.82
0
-1.85
-1.70
-3.02
0
-3.45
0.73
-0.85
-3.34
0
-1.54
-3.87
-2.66
-1.07
42 Markets 35 (83%) losses 7 (17%) wins
-£74.0 £11.31
If you conducted the same experiment but as two experiments side by side but with one always backing with your file and the other always laying with the same file I suspect they'll both deliver losses. The centre of the Bell Curve is not zero - your file is buying the spread multiple times in each market.
-
bastuncriochnatha
- Posts: 35
- Joined: Thu Nov 29, 2018 12:57 pm
sa7med Thanks for that.
Point taken, of course, about sample size. Nevertheless, within the sample limitation the deviation is close to extreme. I have seen this happen consistently for the past 6 months over a much larger sample size.
You say: "because of adverse selection, when offering a price at random, your money will more likely be taken when the market is moving against you "
Not sure about this. Offering a random price will, of course, need a tick against in order to be taken. As you say, a market move against is required in order for the bet to be taken, but after that the frequency of continuation of movement against is surely uncommon as reflected in the 83% loss rate. It indicates a preponderance of "market moving against" states. When all is said and done, if a trader has a view on the direction a market is going to take and then backs or lays that view by offering a position it will need a tick against trader opinion for the offer to be taken. This will happen many times and then might be followed, a reasonable number of times, by the movement predicted by the trader. On a random basis maybe one could expect it to happen more than the 17% win rate would suggest. Fact is, it has to happen in a Make Market scalping success or indeed scalping in general.
The different tick sizes? Not sure what difference that makes in the overall observation of win/loss markets. However, I will look at the phenomenon in future jointly in terms of -/+ ticks and win/loss.
Jukebox. Thanks.
You say: "same experiment but as two experiments side by side but with one always backing with your file and the other always laying with the same file I suspect they'll both deliver losses"
Not sure what you mean by file in this context. My protocol, I will assume. Buying the spread multiple times? Yes, the purchase of the spread, as you put it, does lead to losses when that spread is against the opening position as frequently as 83% suggests. However, I will try what you suggest but do not see how "side by side" can be accomplished unless you mean that I should apply it simultaneously by Laying on one market and Backing on another market within same race? I am not sure what the results of that would indicate.
Bottom line, as I see it and from literal observation, is that a random Lay or Bet results in an adverse move against that opened position far more frequently than makes any sense whatsover.
Cheers, onwards to la-la land
Point taken, of course, about sample size. Nevertheless, within the sample limitation the deviation is close to extreme. I have seen this happen consistently for the past 6 months over a much larger sample size.
You say: "because of adverse selection, when offering a price at random, your money will more likely be taken when the market is moving against you "
Not sure about this. Offering a random price will, of course, need a tick against in order to be taken. As you say, a market move against is required in order for the bet to be taken, but after that the frequency of continuation of movement against is surely uncommon as reflected in the 83% loss rate. It indicates a preponderance of "market moving against" states. When all is said and done, if a trader has a view on the direction a market is going to take and then backs or lays that view by offering a position it will need a tick against trader opinion for the offer to be taken. This will happen many times and then might be followed, a reasonable number of times, by the movement predicted by the trader. On a random basis maybe one could expect it to happen more than the 17% win rate would suggest. Fact is, it has to happen in a Make Market scalping success or indeed scalping in general.
The different tick sizes? Not sure what difference that makes in the overall observation of win/loss markets. However, I will look at the phenomenon in future jointly in terms of -/+ ticks and win/loss.
Jukebox. Thanks.
You say: "same experiment but as two experiments side by side but with one always backing with your file and the other always laying with the same file I suspect they'll both deliver losses"
Not sure what you mean by file in this context. My protocol, I will assume. Buying the spread multiple times? Yes, the purchase of the spread, as you put it, does lead to losses when that spread is against the opening position as frequently as 83% suggests. However, I will try what you suggest but do not see how "side by side" can be accomplished unless you mean that I should apply it simultaneously by Laying on one market and Backing on another market within same race? I am not sure what the results of that would indicate.
Bottom line, as I see it and from literal observation, is that a random Lay or Bet results in an adverse move against that opened position far more frequently than makes any sense whatsover.
Cheers, onwards to la-la land
I meant either using two accounts or two Practice Mode sessions working the same markets with one session following only your lay and close protocol and the other following only your back and close protocol. If, as I suspect they will be, they are both losers it would show that suffering the spread again and again is the issue and not some low probability streak of bad luck that weights your results so far to the negative.bastuncriochnatha wrote: ↑Sun Jan 26, 2020 5:32 pmHowever, I will try what you suggest but do not see how "side by side" can be accomplished unless you mean that I should apply it simultaneously by Laying on one market and Backing on another market within same race? I am not sure what the results of that would indicate.
-
bastuncriochnatha
- Posts: 35
- Joined: Thu Nov 29, 2018 12:57 pm
With respect ...are you seriously suggesting that if I have a situation setup (a given market), running on seperate Practice sessions (Bet Angel/Other Software) where I can take an action(s) and it's opposite ie. a Lay and Bet at the same time (instantaneousely) then both of these are going to produce a consistent loss profile! Really? You are surely suggesting there that anything that is done on a market produces a loss!!!
I am hoping that I have completely misunderstood your hypotheses
I am hoping that I have completely misunderstood your hypotheses
With equal respect, but fewer exlamation marks, your protocol produces a marked loss and expressed surprise that it does so consistently as if it was some statistical anomoly (last time you had a similar problem you reported on this forum that you tried to flip 10 heads in a row). I am merely suggesting a way to demonstrate to your own satisfaction that your protocol is the issue.
-
bastuncriochnatha
- Posts: 35
- Joined: Thu Nov 29, 2018 12:57 pm
Well..... I am at a loss and I apologise for casting doubt on your conjecture.
I did what you suggested.
Fired up other software and BA both in Practice Mode.
Instance 1 of a Market was on Other software and Instance 2 of that market on BA.
1.Decided that a randomly derived Back or Lay would be Offered on one Instance immediately followed by the opposite Offer on the other Instance.
2. The first Offer (whether Back or Lay) that was taken would be immediately followed by Taking price on the other Instance.
3. 30secs count and Green or Red up both Instances.
4. Repeat this from 100K matched into 0 count.
The upshot was as you thought. A stong bias towards loss!
If one instance profited by, say, +£1.00 the other Instance would show a loss of, say, -£2.00.
There was preponderance of the likes of:
-£0.50 -£1.50
£0 -£0.80....
(mainly Red Red)
The occasional green on both would look like:
£2.00 £1.00 (green green only happened once in 10 races)
I cannot get my poor old head around what you call "suffering the spread again and again".
Every interaction with the Instances is randomised.
If a Lay Offer is taken on the first Instance then the Bet price is Taken on the other Instance....
AND VICA VERSA. It can be one or the other condition, And so on!
I am mystified as to why this should bias towards -ve.
Would you be kind enough to elaborate a little on "suffering the spread again and again" for a poor old dullard. My intuition is that this spread in a randomised situation should be both suffered and enjoyed ie now and again in favour of the action...and overall tending to bias results at least reasonabley towards 50/50.
Next up will be a single STEP 1 (above) per setup. That, of course, should pan out approaching 50/50 with increasing attempts.
Do forgive the surfeit of exclamation marks. The upshot of extreme puzzlement.
I did what you suggested.
Fired up other software and BA both in Practice Mode.
Instance 1 of a Market was on Other software and Instance 2 of that market on BA.
1.Decided that a randomly derived Back or Lay would be Offered on one Instance immediately followed by the opposite Offer on the other Instance.
2. The first Offer (whether Back or Lay) that was taken would be immediately followed by Taking price on the other Instance.
3. 30secs count and Green or Red up both Instances.
4. Repeat this from 100K matched into 0 count.
The upshot was as you thought. A stong bias towards loss!
If one instance profited by, say, +£1.00 the other Instance would show a loss of, say, -£2.00.
There was preponderance of the likes of:
-£0.50 -£1.50
£0 -£0.80....
(mainly Red Red)
The occasional green on both would look like:
£2.00 £1.00 (green green only happened once in 10 races)
I cannot get my poor old head around what you call "suffering the spread again and again".
Every interaction with the Instances is randomised.
If a Lay Offer is taken on the first Instance then the Bet price is Taken on the other Instance....
AND VICA VERSA. It can be one or the other condition, And so on!
I am mystified as to why this should bias towards -ve.
Would you be kind enough to elaborate a little on "suffering the spread again and again" for a poor old dullard. My intuition is that this spread in a randomised situation should be both suffered and enjoyed ie now and again in favour of the action...and overall tending to bias results at least reasonabley towards 50/50.
Next up will be a single STEP 1 (above) per setup. That, of course, should pan out approaching 50/50 with increasing attempts.
Do forgive the surfeit of exclamation marks. The upshot of extreme puzzlement.
-
AlexisMartin
- Posts: 60
- Joined: Thu Oct 24, 2019 8:26 pm
how many days have you been running this experiment?
Just one day or over a period of time or is this a litmus test of one day? Have you picked days when the volume is below average or above average? or is it just a random one day experiment hoping to get a normal distriubtion from percieved randomness?
try another controlled group alongside your precieved random variable group so have set of rules eg. only scalp fav's above 5 in 6 runner fields vs your current set, you might find a normal distribuiton in the group where you put more thought into it!
If you are lookng for a normal distrubution in randomeness you will eventually find it then I suugest getting a bank to back you for that professionally!
Just one day or over a period of time or is this a litmus test of one day? Have you picked days when the volume is below average or above average? or is it just a random one day experiment hoping to get a normal distriubtion from percieved randomness?
try another controlled group alongside your precieved random variable group so have set of rules eg. only scalp fav's above 5 in 6 runner fields vs your current set, you might find a normal distribuiton in the group where you put more thought into it!
If you are lookng for a normal distrubution in randomeness you will eventually find it then I suugest getting a bank to back you for that professionally!
-
ruthlessimon
- Posts: 2161
- Joined: Wed Mar 23, 2016 3:54 pm
Then it looks like a perfect candidate for your request. The OP has laid out his procedure and consistently loses.
viewtopic.php?f=42&t=5322&hilit=losing
To my eye it looks as I would expect - a run-of-the-mill bell curve around an amount I would say is entirely consistent with opening a position on a favorite half a dozen times for no particular reason with £20 and closing just a few seconds later and mostly paying the spread for no movement.
-
bastuncriochnatha
- Posts: 35
- Joined: Thu Nov 29, 2018 12:57 pm
Thanks AlexisMartin.
"how many days have you been running this experiment?"
Over 6 months approx 1 day per week. Variations of the protocol described at start of this post. No matter what was tried gave invariably a strong loss bias.
I did say that I had no pretensions towards being statistically competent and just looking for advice on what I was seeing.
JUKEBOX was good enough to tell me that I was "suffering the spread again and again" , whatever that means, due to protocol deficiency which has left me none the wiser, I'm afraid.
"Have you picked days when the volume is below average or above average? "
As stated, will apply protocol to FAV from 100K matched into 0 count (official race start)
"If you are lookng for a normal distrubution in randomeness you will eventually find it"
Really! See https://www.isixsigma.com/tools-templat ... and-tools/ . "Normally distributed data is a commonly misunderstood concept. Some people believe that all data collected and used for analysis must be distributed normally. But normal distribution does not happen as often as people think."
Of course, I would be hoping that in this case, with enough tests, I would start to see a reversion to mean (trend towards normal dist). Even over a short day-per-week-6-months data set. Hence my intention as stated above to restrict the process to a randomly derived Back or Lay Offered on one Instance immediately followed by the opposite Offer on the other Instance which should adhere to Pete Webbs's suggestion of a true random approach to Bet/Lay and produce an overall P/L approaching zero (and so invalidate my protocol),
"I suugest(sic) getting a bank to back you for that professionally!
Ah! You're killin' me you devil you
Anyway, enough is enough and sure the Twilight Zone is not the worst of places to exist and I thank you all (as Freddie Mercury said).
ANYHOO!
I HOPE YOU WILL ALL TUNE IN ON 22 FEB to see the GREAT FIGHTER, TYSON FURY fulfill his destiny to become UNDISPUTED HEAVYWEIGHT CHAMPION and provide further validation to us of the Travelling Community...if any was needed!
"how many days have you been running this experiment?"
Over 6 months approx 1 day per week. Variations of the protocol described at start of this post. No matter what was tried gave invariably a strong loss bias.
I did say that I had no pretensions towards being statistically competent and just looking for advice on what I was seeing.
JUKEBOX was good enough to tell me that I was "suffering the spread again and again" , whatever that means, due to protocol deficiency which has left me none the wiser, I'm afraid.
"Have you picked days when the volume is below average or above average? "
As stated, will apply protocol to FAV from 100K matched into 0 count (official race start)
"If you are lookng for a normal distrubution in randomeness you will eventually find it"
Really! See https://www.isixsigma.com/tools-templat ... and-tools/ . "Normally distributed data is a commonly misunderstood concept. Some people believe that all data collected and used for analysis must be distributed normally. But normal distribution does not happen as often as people think."
Of course, I would be hoping that in this case, with enough tests, I would start to see a reversion to mean (trend towards normal dist). Even over a short day-per-week-6-months data set. Hence my intention as stated above to restrict the process to a randomly derived Back or Lay Offered on one Instance immediately followed by the opposite Offer on the other Instance which should adhere to Pete Webbs's suggestion of a true random approach to Bet/Lay and produce an overall P/L approaching zero (and so invalidate my protocol),
"I suugest(sic) getting a bank to back you for that professionally!
Ah! You're killin' me you devil you
Anyway, enough is enough and sure the Twilight Zone is not the worst of places to exist and I thank you all (as Freddie Mercury said).
ANYHOO!
I HOPE YOU WILL ALL TUNE IN ON 22 FEB to see the GREAT FIGHTER, TYSON FURY fulfill his destiny to become UNDISPUTED HEAVYWEIGHT CHAMPION and provide further validation to us of the Travelling Community...if any was needed!
Don't give up on it. There's nothing wrong in putting such an assertion to the test - and Peter (Euler) has noticed your thread.
The 'spread' I was refering to was the difference between the available back and lay prices on offer. Its at its most obvious when you see a typical race pre off early on with the back book offering 110% and the lay book offering 90% and gradually that gap closes as the race nears start time to around 101% to 99% as the gaps between the offers on both sides closes to within a tick on most selections.
If we take a typical favorite at the stage you are getting involved - with the available back price of say 3.5 (28.57%) and lay price of 3.55 (28.16%) there is a difference of nearly 1/2 a percent that has to be overcome just to break even and before you can profit, regardless of which side you open on. - If you close trades after a few seconds based on time alone you will increase the frequency that you will take the loss of this difference.
Ruthlessimon's graph of your data shows that the market movement is not being particularly unfair with the direction it takes in your 42 races - its just that it is not clustered around zero like you expected. It's my guess that its clustered around that -1.30 because of closing half a dozen times at the best available price and many of those times when no movement happened because of your 10 second timer.
The 'spread' I was refering to was the difference between the available back and lay prices on offer. Its at its most obvious when you see a typical race pre off early on with the back book offering 110% and the lay book offering 90% and gradually that gap closes as the race nears start time to around 101% to 99% as the gaps between the offers on both sides closes to within a tick on most selections.
If we take a typical favorite at the stage you are getting involved - with the available back price of say 3.5 (28.57%) and lay price of 3.55 (28.16%) there is a difference of nearly 1/2 a percent that has to be overcome just to break even and before you can profit, regardless of which side you open on. - If you close trades after a few seconds based on time alone you will increase the frequency that you will take the loss of this difference.
Ruthlessimon's graph of your data shows that the market movement is not being particularly unfair with the direction it takes in your 42 races - its just that it is not clustered around zero like you expected. It's my guess that its clustered around that -1.30 because of closing half a dozen times at the best available price and many of those times when no movement happened because of your 10 second timer.