Probability based trading

[stutch]

Hi,
I wondered if anyone traded blind based on probability.
I have come across an anomoly based on the probability densitiy funcitons of currencies and to put it simply I don't know how to trade it.

A spread (bid/ask) of greater than 1.3 kills it dead, it becomes a losing system.
I normally trade a spread bet account so that is out. (3 pips).

Anything greater than 1.3 shifts it out of line.
Costs of trading are not so important. The spread is very important.

I don't want to publish what I am up to for obvious reasons, but at the same time I am frustrated because I cannot do anything with this.

I know this is unreasonable but if I trade without a spread it returns 50,000 pips on EURUSD in 6 months.
A spread of 1 brings this down to around 1,000 pips. (Probability shift).
I am also using 'indicative' data from dukascopy so it could be utter rubbish.
I have also tried this same algorithm on USD/JPY and it works, it works on EUR/JPY.
I have a basic proof mathematically and experimentally this works.

Because it is probability based I can trade it backwards in time too and it still works.
It works if traded every minute, every 10 minutes, every hour or once per day or completely randomly.
But obviously the PnL is different.
Market direction is unimportant. I think it may apply to other underlying instruments too.
It is risk neutral most of the time (short and long).

Another problem is I can't afford to trade futures, I could not take the short term draw down and I simply don't have the money or experience to trade futures.
Any ideas would be appreciated. This is so frustrating.
( I am not looking to sell this, I may publish the results if I get nowhere trading as it is amusing, Taleb looks for black swans and I think I have found a sky full of pidgeons ).

 

[JonnyT]

The only option is EURUSD Globex future which normally has a spread of 1 but this does vary especially when the US Market isn't open.

However you will still experience slippage.

I would also suggest that you cannot backtest such a slim system on Dukascopy data as does it include just the bid prices, just the offer prices or a combination of both for example?

If you send me a PM I can send you some EUR Globex futures data for you to ponder over. This can be BID, OFFER, TRADES or BID/OFFER combined. Personally I would stick to BID or Offer and make assumptions on the spread / slippage etc.

I still doubt you have a valid system.

Please remember you will lose 0.5 on the spread for commission trading futures aswell.

JonnyT

 

[Tuffty]

If it can be done lengthening the time period of the trade will reduce dealing costs in comparison to market moves.

 

[stutch]

If it can be done lengthening the time period of the trade will reduce dealing costs in comparison to market moves.

Unfortunately time is crucial to this.
Extending the time frame turns this to mush.

Interestingly though in the longer time frame there are some huge gains to be made, with huge risk and no way of 'predicting' the outcome.

This lead to the reasoning behind some of the discrepancies you see posted here in the trading techniques used.

The longer time frame is a trend followers dream, the shorter for the scalper.
Accordingly the limits/stops in one are reversed in the other if you want higher probability trades.

 

[turtle trader]

Unfortunately time is crucial to this.
Extending the time frame turns this to mush.

In which case it's a dead duck! As JT says the narrowest spread you'll find is on the Futures, which is normally 1 pip. If that spread reduces your profit from 50000 pips to 1000 pips, then the 1 pip slippage you'll need to account for leaves you at - 48000 pips & the commission (say 1/4 pip for round figures sake) will give you a total loss of about 60500 pips (if i've understood what your saying correctly).

Good luck though
Simon

 

[stutch]

Thanks to JohnnyT for comments and assitance.

I can confirm that the dukascopy EUR/USD data and the EURGLOBEX futures data (bid) has a correlation of 0.98 for a sample size of 1918 elements. This means it is 'almost' the same or there is a very close correlation of open prices of the 10 minute data and a sample taken from the EURGLOBEX at 10 minute intervals.

I have attached a chart the EURGLOBEX is in pink, EUR/USD in blue.

It is interersting that using the same alogrithms on dukacopy eur/usd data and eurglobex I get exactly the same results.
I have attached a couple of jpg's that show the impact of a 1 pip and 2 pip spread for the period I have data. I note this is too short a period for conclusive proof.
Pips up the side, samples at 10 min interval along the bottom.
The PNL curve is adjustable, if the draw down is too much, don't trade so often!
Proftit will be lower too.

The reason for the difference is attributable to the shift in probability relationships caused by the spread. That is to say if you are long and you have a limit order in place, you have obtain a price of limit + spread. The probability of this occurring is less than the limit alone.

 

[charliechan]

Hi,
I wondered if anyone traded blind based on probability.
I have come across an anomoly based on the probability densitiy funcitons of currencies and to put it simply I don't know how to trade it.

A spread (bid/ask) of greater than 1.3 kills it dead, it becomes a losing system.
I normally trade a spread bet account so that is out. (3 pips).

Anything greater than 1.3 shifts it out of line.
Costs of trading are not so important. The spread is very important.

I don't want to publish what I am up to for obvious reasons, but at the same time I am frustrated because I cannot do anything with this.

.....

so let me get this straight. you have a system, but you dont know how to trade it. also, you dont want to tell us anything about it either.

if i were a police man, id arrest you for wasting time.

 

[Directional]

if your system is a super-scalper then you have to evaluate seriously if it is workable in the market with such "thin" average profit per trade.

The spread and/or commissions & slippage are just one of the costs of doing business in trading directionally. If you do not properly account for them in your system design, then unfortunately you may be wasting your time on the system as it is, without looking at ways to increase the value of the average trade to ensure it is profitable in real trading including these costs.

Does your system use market or limit orders for entry? how many trades did it take over your sample?

 

[stutch]

Does your system use market or limit orders for entry? how many trades did it take over your sample?

Points noted.

To start answering your question :

I do not know what the market will do next. If I do not know what it does next then my entry point is not important. The entry point for the graphs is the open price on a 10 minute interval, but I could choose any point that exists within that bar to open a long and a short position.

What happens next is up to the market. I set a limit and a stop and leave it to expire on either the limit or the stop. For the graphs shown I have used a 1 pip or 2 pip spread so if the limit + 1 is hit I record only the limit. If the stop is hit then I record the stop + 1 and so on.

This gives me a market neutral position until one of the orders is hit and I take a profit.
Then I leave the other to be stopped out or hit the limit if it mean reverts.
( Please don't try this with money unless you fully understand what you are doing.
I do not use money on this myself, its a research project for me. )

I don't care what the market has done, I do not care what it does next.
There is no hindsight involved, no patterns, no stochastics, no candles or bars.

To answer the question about number of trades per hour/minute or day it does not matter.
If you imagine this chart was on a rubber sheet and you pulled the ends the time axis would extend and the profit and loss would move in towards the axis. That is what happens if you trade every hour. If you trade ever minute then pull the rubber sheet vertically and the PNL expands with it and the time frame shrinks. Note : Losses increase too and you'd suffer the gamblers curse.

So if you have a controllable PNL with a limited risk position you can see why I am researching this.

 

[stutch]

so let me get this straight. you have a system, but you dont know how to trade it. also, you dont want to tell us anything about it either.

if i were a police man, id arrest you for wasting time.

I apologise for not being explicit in my articles but I do not want anyone to treat what I say as advice or guidance on trading. I see this as the start of some research work and I don't really want to publish what I am doing at this stage or my work becomes worthless.
My original question was whether anyone else used this method, I was enquiring to see whether my thoughts were original. I did not want to appear to gloat.

I will put more substance in to one of the more constructive replies if that is what people want.

 

[Directional]

The entry point for the graphs is the open price on a 10 minute interval, but I could choose any point that exists within that bar to open a long and a short position.

What happens next is up to the market. I set a limit and a stop and leave it to expire on either the limit or the stop. For the graphs shown I have used a 1 pip or 2 pip spread so if the limit + 1 is hit I record only the limit. If the stop is hit then I record the stop + 1 and so on.

This gives me a market neutral position until one of the orders is hit and I take a profit.
Then I leave the other to be stopped out or hit the limit if it mean reverts.
( Please don't try this with money unless you fully understand what you are doing.
I do not use money on this myself, its a research project for me. )

Interesting. so you are essentially using an element of spreading by being both long and short at the same time, and looking for a price point to remove one side or the other of your hedge.

This can be done in forex - simultaneous long & short, but not in futures, unless you use the next expiry future contract month.

Although - in forex, by having both long and short in the same currency spot, you are essentially flat, but paying the spread for both positions. Would your approach be "invertable" so that rather than being "flat" but having two positions on - you instead dont take any exposure until price hits the predetermined point at which you would have taken one side of the hedge off and created exposure - use that point to take a straight position to replicate what you were trying to acheive, but cheaper since you are only taking one position or the other. if you follow.

 

[stutch]

Does your system use market or limit orders for entry? how many trades did it take over your sample?

One point I forgot to add is that not all positions are 50/50 positions. As a scalping strategy if your limits are closer to your opening position and your stops a further away you move the odds in your favour.
This is contrary to what a lot of books say so let me prove it.

Take a limit order of 10 pips and a stop of 50 for example. It would be reasonable to assume the odds are 50/(10+50) = 5/6 and 1/6 . You can never have greater than 1.

However the mean time to 10 pips is around 30 minutes on EURUSD and around 80 for 50 pips. This moves the odds towards the 10 pips. Only if the mean time was the same at say 30/30 would the odds remain 5/6 and 1/6 respectively. This means the probability of a 10 pip limit hit is around 0.86 and 50 around 0.14 this gives a geometric mean of around 0.2 and hence profitability.
Note: This only applies to the short term and does not take in to account spread, costs etc.

Only a trend follower should keep his stops closer than his limits.
I can prove that too but I don't want to appear to be trying to show off and its not relevant to the original article.

Again I am not offering advice on what to do, there are far more factors than math to consider.
I am not a good mathematician or trader so be warned.

 

[stutch]

Although - in forex, by having both long and short in the same currency spot, you are essentially flat, but paying the spread for both positions. Would your approach be "invertible" so that rather than being "flat" but having two positions on - you instead dont take any exposure until price hits the predetermined point at which you would have taken one side of the hedge off and created exposure - use that point to take a straight position to replicate what you were trying to achieve, but cheaper since you are only taking one position or the other. if you follow.

If you look at the long term, taking long positions is profitable but you may have a severe draw down. If you only took short positions, that is profitable too.
The shorts are more often profitable when the longs are losses and vice versa.
It combines two winning systems to reduce draw down.
It is like a Brownian ratchet.

That is why the limit + spread kills this if the spread is too great.
It is like walking up the down escalator.

I have not researched invertible positions properly.
Initial trials indicated this 'capture' system would work and I can explain why so I have focused on this 'capture' type system.
My initial trials with invertible positions were inconclusive and I do this in my spare time so I have to focus on one thing at a time. There key dependency is mean reversion and it can be made to work. I had tried it on AZN some time ago with a little success but it depended on the nature of the equity, volatility etc.

I will bear your comments in mind and I do appreciate the thoughts of the community here.

 

[growltiger]

I apologise for not being explicit in my articles but I do not want anyone to treat what I say as advice or guidance on trading. I see this as the start of some research work and I don't really want to publish what I am doing at this stage or my work becomes worthless.
My original question was whether anyone else used this method, I was enquiring to see whether my thoughts were original. I did not want to appear to gloat.

I will put more substance in to one of the more constructive replies if that is what people want.

I think this method (run pairs of conracts, realise profits and allow unrealised losses) has made a large number of appearances in various guises.

There is a recent Elite Trader thread where the contributor ElectricSavant proposes a system for trading on these lines:
http://www.elitetrader.com/vb/showthread.php?s=&threadid=47844

 

[bodemloze_put]

Take a limit order of 10 pips and a stop of 50 for example. It would be reasonable to assume the odds are 50/(10+50) = 5/6 and 1/6 . You can never have greater than 1.

However the mean time to 10 pips is around 30 minutes on EURUSD and around 80 for 50 pips. This moves the odds towards the 10 pips. Only if the mean time was the same at say 30/30 would the odds remain 5/6 and 1/6 respectively. This means the probability of a 10 pip limit hit is around 0.86 and 50 around 0.14 this gives a geometric mean of around 0.2 and hence profitability.
Note: This only applies to the short term and does not take in to account spread, costs etc.

This means that on average you would make (0.86*10)+(0.14*-50)=1.6 pips per trade. If you take a 1 pip spread on both sides you end up with a 0.4 pip loss. Please remember that markets change over time. Not all historic probabilities can be transformed to the future.

BP

 

[nicholaf]

Interesting. so you are essentially using an element of spreading by being both long and short at the same time, and looking for a price point to remove one side or the other of your hedge.

This can be done in forex - simultaneous long & short, but not in futures, unless you use the next expiry future contract month.

Although - in forex, by having both long and short in the same currency spot, you are essentially flat, but paying the spread for both positions. Would your approach be "invertable" so that rather than being "flat" but having two positions on - you instead dont take any exposure until price hits the predetermined point at which you would have taken one side of the hedge off and created exposure - use that point to take a straight position to replicate what you were trying to acheive, but cheaper since you are only taking one position or the other. if you follow.

good idea, but too bad, the real world is different. the spreads are too large and with trans cost and slippage, you're basically dying by 1000 cuts.

this is the prob with many backtested systems. misses the real world factor and this is deadly.