### Author Topic: Basics Of A Quantitative Roulette Strategy  (Read 4708 times)

0 Members and 2 Guests are viewing this topic.

#### fiben7

##### Basics Of A Quantitative Roulette Strategy
« on: May 16, 2019, 11:26:20 PM »

Hi dear roulette enthusiasts,
Given my professional experience in quantitative trading for the financial markets along with my deep interest in the roulette game and challenge, I hereby post some basic thoughts of mine for your consideration, analysis and discussion based on the forum's general high level of experience and expertise.

1. What type of roulette

I believe that the most robust and backtestable (backtesting is a common concept in algoritmic quantitative trading and easily applicable to roullete dynamics, please google) roullete type is the automatic real mechanical roulette wheel. In other words, the roulette wheel that you can actually see it yourself live at a casino venue however there is no human dealer involved  and the spin is originated automatically by a specified mechanism. In such a roulette wheel, the statistical phenomenon whose dynamics the player is trying to predict and thus generate a positive performance are most controlled and unbiased.

The human dealer roulette could be hugely biased by the dealer's style and "hand" and since after a short time, dealers change and new ones come into the scene of the same roulette wheel, the general environment of the experiment is unstable and prone to biases.

Regarding the electronic roulette wheels where numbers sequences are based on computer random generators etc, I believe that by no means should any educated player trust the results of a strategy based on such number sequences; sophisticated algorithms tend to suspiciously favour winnings just at the right time only for players to feel good enough to visit a real casino and face the hard truth of real statistical anomalies under a well structured chaos.

2. The amount of bet

I believe that the fixed bet per spin is the best way to test and analyze if your betting strategy and logic is robust enough. The double on loss, double on gain, fibonacci sequence and all other more complex progression styles of betting may be helpful sometimes in optimising  the risk/return ratio of the strategy but they do not and should not make a real difference in the dynamics of the strategy's perfomance. The real difference is made by the prediction of the next spin itself. To rephrase, a strategy must first be robust and successful enough in the prediction side of things (red/black, odd/even, high/low, dozens, voisins/tiers/orphelins, straight numbers etc.) whereas the betting amount and any progression should be a risk management tool just to enhance the risk/return characteristics of the strategy - and not to turn a losing strategy into a winnig strategy. Thus, using a fixed bet on each spin, adds credibility and focuses on the event prediction side of things.

3. The type of bet

In general, roulette offers by default some "50-50" events (red/black, odd/even, high/low) and of same concept bets (sixes, dozens etc) as well as more sophisticated clustered neighborhood bets like voisins/tiers/orphelins; there are also straight number bets.

Which type of bet a playes chooses has to do with the expected payout profile of the strategy, i.e. bet 6 chips on Tiers chips to get 18, bet 1 chip in Black to gain 2 etc. and the general risk profile of the player.

Nevertheless, in any type of bet the strategy is based upon, there should be a simple general rule, just like in the financial markets trading strategies.

Either

Follow the Trend, i.e. if the previous spin is Black (Voisins) , then bet again on Black (Voisins)

Or

Follow the Mean Reversion, i.e. if the previous spin is Black (Voisins) , then bet on the Red (Tiers).

For more sophisticated strategies, there could be an internal indicator that locally turns the strategy from Trend to Mean Reversion and vice versa.

4. Budget and risk metrics

Once the strategy is developed, it should have simple rules that need to be followed strictly under the short amount of time that each spin follows after the previous one. Given the time constraints and the necessity of simplicity, we need to quantify the performance of our strategy by simple statistics.

First we need to state our budget (this will be our initial portfolio value on which we will translate the cash profit or loss into meaningful returns). Lets say there is a cash budget of EUR 500.

Then, given that we have saved the real data of real spins, we can backtest and check the profit and loss behavior of our strategy over time and the data of spins. This exercise generates the most useful statistics for our analysis.

For example

The hit ratio of our strategy; how many successful predictions versus the total spins. This is very useful on the risk management side. For example if a strategy has an overall 25% hit ratio and for some reason during the last 20 spins, the strategy performs a 60% hit ratio temporarily, then it would be safe (than sorry) to expect more prediction misses at the next 20 spins, so that the hit ratio level drops to its normality of 25% (in such a case a player could take the profit and bet no more till the hit ratio resumes normality; and thus saving a lot of lost bets).

The average/max/min net profit on winning spins

The average/max/min net loss on losing spins

The cumulative cash (or return) profit and loss performance over the totality of spins (this creates the most important timeseries for analysis, many many financial statistical tools could be used to decode the strategy dynamics here).

The maximun drawdown of the strategy; how much cash or return % the strategy would lose historically should the player enter and leave the wheel at the worst possible time (spins) window possible. For example if the maximum drawdown in cash of a strategy across 10,000 spins is EUR 340, then a player would feel at more ease should the strategy temporarily is costing a EUR 200 loss as the player knows that this is an event statistically possible and not abnormal whereas if the player is down EUR 400, then something looks very suspicious as the previous historical worst case scenario max loss is breached by EUR 60 (400-340).

I could go on and on such risk/return metrics though I am sure you got the point.

5. The Mystery number 0 and neighbourhood Orphelins

I believe that what makes the difference in a robust roulette strategy is the way the strategy deals with the number 0 (on classic bet types like Black/Red etc) and the Orphelins (on classic Voisins/Tiers etc).

A sophisticated (fixed or dynamic) logic to either  include/exclude 0 or Orphelins is what makes the risk/return ratio tilt to the favor of the player or not.

6. Reality

I have spent quite of few hours testing various strategies in real venues on automated real mechanic roulette wheels, most importantly gathering data so that I can get then back home and backtest new and old ideas in search for smooth, robust positive perfromance. Screening between strategies becomes more and more easy as new real data come into place; the surviving strategies depict a resistance in bad temporal spins sequences and are consistently profitable in favourable ones.

No roulette strategy has 100% hit ratio, all strategies have drawdowns and temporal losses - that should be recovered back under a fair amount of time - and most importantly: leave psychology aside and just follow the strategy's rules.

That was that from me for now, looking forward to your feedback and discussion further.

Best of luck across!
Konstantinos
« Last Edit: May 16, 2019, 11:42:16 PM by kav »

The following users thanked this post: kav

#### Third

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #1 on: May 16, 2019, 11:33:41 PM »
Please edit this post.  Place the cursor in the typing field, hit CTRL-A and then hit the button that looks like this:

The following users thanked this post: GIAJJENNO, UnlikelySam, fiben7

#### kav

• Hero Member
• Posts: 2350
• Thanked: 1320 times
• Gender:
##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #2 on: May 16, 2019, 11:53:20 PM »
Hello fiben,

How you increase or decrease your bets can make a world of difference, see here:
Does A Series Of Short Sessions Combine To Make A Long Series?
Also see here:
Technical analysis for roulette and here: Trading secrets and strategies

The following users thanked this post: Third, UnlikelySam, fiben7

#### fiben7

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #3 on: May 17, 2019, 06:17:53 AM »
Dear kav,

Thank you for your feedback and very nice to meet you here.

Sure, how to increase/decrease the bets can make a difference; my point for the fellow strategy makers was that instead of focusing on this aspect, we should primarily focus on the prediction aspect.

First ensure the strategy is sucessful enough to predict and follow certain simple or complex statistical patterns and once happy then enhance your risk/return ratio through bet progression.

In other words, if the strategy works with a fixed bet rationale, it can only get better (or of same strength) using bet progression. Though, first get the signal right, at an average hit ratio than combined with your average payout ratio, makes the strategy smooth and positive in the long run!

Best

#### UnlikelySam

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #4 on: May 17, 2019, 07:44:03 AM »
Hi Fiben7 and welcome to the forum...

" In other words, if the strategy works with a fixed bet rationale, it can only get better (or of same strength) using bet progression. "

I take it for granted you are referring to flat betting  and thereafter using a positive bet  progression. If I am wrong please clarify. I agree about focusing on our bet selection more in detail and rather implement a flat betting or positive progression ; however improving one's accuracy of prediction is the most difficult task on hand. With your statistical / financial markets background have you managed to make any breakthroughs in the above?

Regards

Sam...

The following users thanked this post: kav, Third, fiben7

#### dobbelsteen

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #5 on: May 17, 2019, 08:29:46 AM »
Fiben 7: In your paper I recognize a lot of my theory. You make no difference between a strategy and a system.  Systems have fixed rules and can be programmed a strategy does not. A play session consists of a lot of small particular sessions. Particular sessions are together not a long random session. One of the most important rule of a strategy is Hit and Run.

The following users thanked this post: Third, UnlikelySam, fiben7

#### kav

• Hero Member
• Posts: 2350
• Thanked: 1320 times
• Gender:
##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #6 on: May 17, 2019, 09:11:58 AM »
IF we divided a strategy on WHAT to bet and HOW (much) to bet.
Then IMHO the HOW is much more important and the key to win.
I have never seen a bet selection that would produce an advantage simply by flat betting.

The following users thanked this post: Third, Astutillo, fiben7

#### fiben7

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #7 on: May 17, 2019, 09:17:16 AM »
Hi UnlikelySam, thanks for the welcoming, nice to meet you!

Confirmed, agreed, I am referring to flat betting and positive bet progression. For example, bet 6  chips to cover the Tiers as per the classic way or 4 chips to play straight the 4 black numbers of of the Tiers. With flat betting your budget is more controlled and is less likely to explode.

Regarding any breakthroughs of my own, yes, I believe I have in hand a couple of strong strategies (or better system as dobbelsteen correctly suggests) and some variations on these. These are based on most of the points in my post, i.e. an effort to recognize trend or mean-reverting processes within the roulette dynamics and combining them in order to achieve a promising risk/return ratio.

For example, is Voisins/Tiers a mean reverting process and if yes, how should you handle the Orphelins? And when betting on the Tiers, how do you handle the black/red, odd/even, the dozens etc?

I think I need many more random visits to the actual automatic mechanic roullete wheel to gather more real data so that I have a very strong case. Once done, I neee ti try the same system on another automatic mechanic roullete of the same casino and then on another automatic mechanic roulette of another casino and so on.

And if the systems performs in a smooth and positive manner through these tests and maintains a positive return with relatively small drawdowns and sticks to the expectes hit ratio, then I could say I have a very very strong case.

The point I have reached so far is extremely promising and hands me over very decent returns with little worries and anxiety when playing.

I will keep you posted!

The following users thanked this post: kav, Third, UnlikelySam

#### fiben7

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #8 on: May 17, 2019, 09:21:43 AM »
Dear dobbelsteen,

Fully agree on your point, the correct wording of what I am proposing is System and not Strategy, one that can be programmed and backtested and whose rules are fully articulated and should be invariant of the session, theal machine, the dealer etc.

I also agree that locally, in particular sessions, there are areas of prediction that could be modelled and exploited.

#### fiben7

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #9 on: May 17, 2019, 09:27:05 AM »
Dear kav,

I see your point and I am not here to disagree, I respect it and give it a lot of thought.

As UnlikelySam says, the prediction part is the most difficult and this is what I am trying to achieve, in an even more constrained manner through flat betting.

I have significant experience in analysing trading systems so in case I succeed in the above, the WHAT to bet, then the HOW (much) to bet should be the easiest part that would optimally risk manage an already winning system.!

Best regards

The following users thanked this post: kav, UnlikelySam

#### dobbelsteen

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #10 on: May 17, 2019, 02:28:08 PM »
fiben 7: The roulette game has 37 different chances. The table has 157 places to bet. All kinds of bets can be brought back to  a number bet .Clusters of numbers on the wheel or table are number bets. The features of the cluster bets are not the same. Fibonacci, D`Alembert  , Martingale etc. are unjustly called systems but they are particular bet selections.

For spreading the risk, you must play more different strategies.

The following users thanked this post: Third, fiben7

#### fiben7

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #11 on: May 17, 2019, 03:15:08 PM »
dobbelsteen: agreed, it all comes back to number bets. The essence is what you bet on each number and what is the payout if the number is hit.

For example, my most prominent system only bets in straight numbers, on average in 9 of them chosen from the system's logic and signal generation and the gross payout is always 36. With a consistent hit ratio of 28% as measured over real data, this is a consistently winning strategy with low cash drawdowns.

Now on spreading the risk, also agreed, this is one of the most importan tools in finance, the so-called diversification. I have tested whether any kind of combinations of my top 5 systems would yield a better overall system, and it surely does.

To speak in numbers, my best system has a "sharpe ratio" of 0.83 whereas the best combination of my top systems has a "sharpe ratio" of 0.95.

The caveat here of course is the execution side of things. Best diversification could come from combining say 10 different systems, however betting within 60 seconds between each spin in 10 different ways could simply become non-feasible.

Thus, I try to keep it all simple and practical, having a diversification of 2  relatively simple systems.

Thanks again, much appreciate your interest!

Best
« Last Edit: May 17, 2019, 03:29:55 PM by fiben7 »

The following users thanked this post: kav, Third

#### MrPerfect.

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #12 on: May 17, 2019, 06:42:29 PM »

All pro quants now are busy making real money. Shorting criptos,  buying AMD...ets. Making kind of money you will never expect to do with roulette... especially with your brilliant ideas.
I love when folks speak " pro experience ".

#### fiben7

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #13 on: May 17, 2019, 07:49:03 PM »
Hi Mr. Perfect,

Sure, pro quants can use models for all kinds of "markets" and price timeseries.

No harm testing a popular and extremely hard to model "market" like the roulette. The amount of money one could make is not the primary reason to research on the topic.

The only reason I noted my professional experience was to better introduce and make sense of my trading approach to the roulette system concept; and not to brag or anything.

By the way, quants and cryptos do not really blend nice together, bad point.

The following users thanked this post: UnlikelySam

#### Third

##### Re: Basics Of A Quantitative Roulette Strategy
« Reply #14 on: May 17, 2019, 10:08:52 PM »
Congraulations on your system.  I wish you the best in testing it and thank you for sharing about it!

The following users thanked this post: fiben7