The ACD Method

Quote from flip:

Looking at the basics of ACD in detail, I noticed that in his book on page 12, Fisher divides the day into 64 five-minute intervals and states that based on random-walk theory the first bar would be the high 1/64 of the time, the low 1/64 of the time and hence the high or low 1/32 of the time. In reality, he states, the opening range tends to be the high or low 17-23 percent of the time which is the edge of this approach.

In order to validate this statement, I used 5min data of the S&P500 future (ES) since Jan 2000, dividing the day into 5min intervals which gives 81 bars per day.

I then calculated the following empirical probabilities for each bar of the day (displayed in the attached chart):

- the high of the day is made until bar number n (blue bars top left)
- the low of the day is made until bar number n (blue bars top right)
- the high and the low are made until bar number n (blue bars bottom left)
- the high or the low are made until bar number n (blue bars bottom right)

I did the same using random-walk data, which is displayed as red line in each chart.

The results are quite interesting, most noticeable:

There's indeed a pretty large difference between the empirical data and random-walk, especially during the beginning of the day: Looking at the bottom right chart you can see that the probability for a high or low during the first period of the day is much larger than random-walk theory would suggest.

However, you can also see that the probabilities for random-walk data are not as low as Fisher suggests:

According to Fisher - when having 81 bars per day - the probability for the first bar of the day to be the high of the day would be 1/81 (~1.23%), same for the low, the probability for either a high or low would then be 1/81 + 1/81 = 2/81 = 2.47%. However, P(H or L) is not simply P(H) + P(L) but instead P(H) + P(L) - P(H and L), where P(H and L) = P(H) * P(L).
On the other hand, the relationship is not linear, i.e. the probability for high of day until first bar = 1/81, until second bar = 2/81, until third bar = 3/81 etc. is not valid. You can see this in the top left and right chart, the red line is not linearly increasing.

So I would conclude that there's indeed a difference between empirical data and theory but not as large as Fisher suggests. Whether this difference is enough to be regarded as an edge is something I cannot say. Based on postings in this thread it seems it's not enough on a stand-alone basis, but definitely useful as a starting point.


Any feedback appreciated! Did anyone look at similar stuff, e.g. as a method for detecting markets with the highest deviation from random-walk?
More...


This is a great thread sans the bickering..take a look...:)

http://www.elitetrader.com/vb/showt...25224&highlight=chart+patterns+in+random+data
 
Quote from flip:

Looking at the basics of ACD in detail, I noticed that in his book on page 12, Fisher divides the day into 64 five-minute intervals and states that based on random-walk theory the first bar would be the high 1/64 of the time, the low 1/64 of the time and hence the high or low 1/32 of the time. In reality, he states, the opening range tends to be the high or low 17-23 percent of the time which is the edge of this approach.

In order to validate this statement, I used 5min data of the S&P500 future (ES) since Jan 2000, dividing the day into 5min intervals which gives 81 bars per day.

I then calculated the following empirical probabilities for each bar of the day (displayed in the attached chart):

- the high of the day is made until bar number n (blue bars top left)
- the low of the day is made until bar number n (blue bars top right)
- the high and the low are made until bar number n (blue bars bottom left)
- the high or the low are made until bar number n (blue bars bottom right)

I did the same using random-walk data, which is displayed as red line in each chart.

The results are quite interesting, most noticeable:

There's indeed a pretty large difference between the empirical data and random-walk, especially during the beginning of the day: Looking at the bottom right chart you can see that the probability for a high or low during the first period of the day is much larger than random-walk theory would suggest.

However, you can also see that the probabilities for random-walk data are not as low as Fisher suggests:

According to Fisher - when having 81 bars per day - the probability for the first bar of the day to be the high of the day would be 1/81 (~1.23%), same for the low, the probability for either a high or low would then be 1/81 + 1/81 = 2/81 = 2.47%. However, P(H or L) is not simply P(H) + P(L) but instead P(H) + P(L) - P(H and L), where P(H and L) = P(H) * P(L).
On the other hand, the relationship is not linear, i.e. the probability for high of day until first bar = 1/81, until second bar = 2/81, until third bar = 3/81 etc. is not valid. You can see this in the top left and right chart, the red line is not linearly increasing.

So I would conclude that there's indeed a difference between empirical data and theory but not as large as Fisher suggests. Whether this difference is enough to be regarded as an edge is something I cannot say. Based on postings in this thread it seems it's not enough on a stand-alone basis, but definitely useful as a starting point.


Any feedback appreciated! Did anyone look at similar stuff, e.g. as a method for detecting markets with the highest deviation from random-walk?
More...

I don't think the edge in the ACD system is derived from the fact that the opening 5 min bar tends to be the high or low more often then randomness would suggest. I think Fisher used that to show that markets are indeed NOT random. It was his thesis paper at Wharton. Form that general idea, that the markets are not random, he created ACD.

The whole idea of believing in trends or breakouts violates the random walk hypothesis. So you can't even have ACD if random walk is valid. So first he set out to prove that markets are not random. Once he did that, he built a methodology around the idea that there are edges in the market.
 
Quote from flip:

I haven't read this thread, but I think it's more interesting (and hopefully lucrative) to find deviations from random walks in real data instead finding trends in artificially generated random walks :)
More...

there is more going in the thread than that, but I know what you are saying. That said if you found a "pattern" in seemingly random data, traded it, made money, bought an island, etc. would it matter? Read thru it when you get time there is some good "food for thought" there. :)
 
Quote from kinggyppo:

Mav my thesis has always been there are trends in the mkt which can be exploited, random or non random is irrelevant to me.
More...

Well, trends can't be exploited if markets are truly random. That was Fisher's point.
 
Quote from Maverick74:

Well, trends can't be exploited if markets are truly random. That was Fisher's point.
More...

I believe he used the term statistically significant, I don't personally believe the mkt or the universe is truly random, but a simple monte carlo of a coin flip will show "trends" ie. 10 heads in a row. Anyhow this is a topic for the other thread. Looking for a touch of 114.20 on spy. :cool:
 
Quote from kinggyppo:

I believe he used the term statistically significant, I don't personally believe the mkt or the universe is truly random, but a simple monte carlo of a coin flip will show "trends" ie. 10 heads in a row. Anyhow this is a topic for the other thread. Looking for a touch of 114.20 on spy. :cool:
More...

Of course trends exist. The debate in the academic community has been whether or not they can be "predicted". I have yet to find anyone who can predict when heads will come up 10 times in a row using a fair coin. :)
 
Feedback for Flip.

Flip. ACD is about RISK management. There is no edge in ACD, other than a viable method to control your risk. ACD is about knowing where to get out more than knowing where to get in. Where to get in after a market makes an A is your business. But, when a market makes an A you are not allowed to trade counterbias unless it makes a C. This alone will keep you from being hacked to bits, which is the fate of too many traders. Death by a thousand cuts. ACD eliminates that.

Myself I am a currency trader, profitable, but my equity swings were HUGE! Up 300% one month, only to give it back over the next three. ACD gives traders with a system a way to qualify their signals. ACD guarantees you lose small and win big. At the end of the day, this is all you need to really learn to be an Elite Trader.

Also ACD, if you already have a system allows you to follow many many more markets. Thanks to ACD I can now follow every pair Oanda offers, and normalize my risk. What that means is that I can wait for exactly what I want price action to do, and not just close enuf. If you have not watched the videos Mav put up, please do that. For me, it was like going from a Bachelors to a PhD. Fish gives traders who are breakeven or slightly profitable the tools to turn the corner.

Seriously, watch those videos. Do not skip anything, and watch them all.
 
You must log in or register to reply here.
Back
Top