what do you guys think about this?

It's easy to understand why this guy has all the beer money he needs. He is a master at sleight of hand, and here's why:

"Curiously, TFP does not predict the CLOSE of the S&P very well but is astonishingly good at predicting the OPEN! This opens up the novel idea of the "Night Trader", run the figures through TFP just before market close to get a prediction of tomorrow's open and take an appropriate position just before market close."

Of course, this is hooey. Why? Because his "projection" of the next day's open is based on the previous day's close, and if you look at his projections, the next day's open is almost always within two points of the previous day's close. The sleight of hand occurs because you're looking at the chart of the Open (predicted = RED, actual = BLUE), and the two lines seem to match perfectly most of the time. Then, you look at the next day's prediction and bang, it appears to nail the open almost every time. But in fact what you really need to do is look at the actual close in the upper left-hand chart and then compare that value with the prediction of the next day's open in the lower right-hand chart. Most of the time, there's hardly any difference.

Here's a sample table I did, incrementally adding each day of data one day at a time to test out the predictions in faux real-time.

(C) = Previous Day's Close
(P) = Predicted Open
Action is SHORT or LONG based on P compared to C
(O) = Actual Open
(N) = Net

01/21/03 (C) 903.10 (P) 902.00 SHORT (O) 904.70 (N) -1.60
01/22/03 (C) 888.50 (P) 890.90 LONG (O) 883.30 (N) -5.20
01/23/03 (C) 877.50 (P) 876.40 SHORT (O) 883.50 (N) -6.00
01/24/03 (C) 883.00 (P) 884.60 LONG (O) 882.00 (N) -1.00
01/27/03 (C) 860.30 (P) 860.20 SHORT (O) 853.50 (N) +6.80
01/28/03 (C) 847.30 (P) 848.50 LONG (O) 852.50 (N) +5.20
01/29/03 (C) 854.50 (P) 860.00 LONG (O) 849.90 (N) -4.60
01/30/03 (C) 860.70 (P) 861.50 LONG (O) 863.50 (N) +2.80
01/31/03 (C) 840.00 (S) 838.40 SHORT (O) 837.00 (N) +3.00

Total net during sample period = -0.60

Of course, the real absurdity of the program is the difference between the close and the predicted open that it generates. Here are the absolute values:

1.10, 2.40, 1.10, 1.60, 0.10, 1.20, 5.50, 0.80, 1.60

Average = 1.71

I'd like to meet the trader who is willing to stake an overnight position based on this average expectation while assuming the overnight risk. What happened in the two cases where the difference between close and predicted open was greater than 2.00, where the program should supposedly have the greatest predictive power? One would have lost -5.20 on the first trade and -4.60 on the second trade. Further, the prediction on 1/27 was just 0.10 off the close, not enough of a difference to trade but yet it turned out to be the biggest trade through pure luck.

Granted, the sample size presented here is small, but it's pretty clear what's happening here. I noticed that the Tomorrow's Financial Pages program was initially written in 1994 and then rolled out again in 2002. Looks like he hooked a new generation of suckers...

Here's your boy:

http://vader.brad.ac.uk/finance/SJShepherd.html
 
The stong form of EMH holds that all that is known or knowable is already reflected in the price of stocks, thus there is no incentive to spend time or money trying to obtain "better" information.

My understanding of the problem with EMH, is that it creates a paradox:

If there is no incentive for anyone to spend time, money,etc
to gain better info and consequently an "upper hand" in/on the market, then the paradox is how the information makes it to the market in the first place---i.e.if no incentive to obtain the data, how is it already reflected in prices?
 
= Exercising My Head

God, I need a drink, now...................Jaguars, here I come.

(By the way I don't think the titty joint realizes that Jaguar, according to the SEC report of Daytrading BD's, is another name for a conversion)
 
The logic in the original argument is both flawed and accurate. As has been pointed out, the emh is accurate when it comes to price almost by definition, available information is reflected in the price of a stock. And yes, linear analysis is a very useful tool as well.

To apply non-linear analysis is really a final step in honing a trade. this can never be done in a system and relies on a traders skill and experience to overlay the linear analysis that cannot predict flaws within itself. Therefore, in order to be successful, the most successful trader is likely armed with good analysis, but must be a trader first a foremost to avoid the pitfalls the systems overlooks.
 
The other issue nobody has pointed out is the the EMH is exactly what it says, "a hypothesis." Therefore, in its simplest sense, it is a best guess starting point for further exploration.
 
Assume the market is efficient. When did that happen? When was the day from then on the market was efficient? 1850, 1930, 1970, 2000, yesterday?
Assume the market is efficient. What is the measure? If I had had the tools I have now back in 1955, could I have made money with it then? If so, is "efficiency" a function of tools at hand? If so, is the market more or less efficient for people with different tools?
Assume the market is efficient. Wouldn't that mean that there is no information- or tool-difference within the group of participants?

I think it is all very easy. Those who have not found a way to profit in the market place, call it efficient. And for themselves, they are perfectly right: they have no hope for a profit, exceeding chance.

The concept as such is so popular because it sounds so scientific, yet is in principle easy to comprehend and has many more people as a proof than successful traders as the contraries (and this last group often prefers silence to noise ...).


peace
 
Gordon Gekko
Elite Member

Registered: Nov 2001
Posts: 1925


01-30-03 02:03 PM
what do you guys think about this?
i'm not smart enough to even understand it really.. for example, after reading the following, i could not in my own words explain it to someone else. lol can someone please explain this in simple terms?

http://vader.brad.ac.uk/finance/tfp.html

specifically, i'd like to understand this part:

>"In spite of the volumes of rubbish in classical economics texts, >the Efficient Market Hypothesis is NOT TRUE. There is now >overwhelming evidence to the contrary.
Sure there is evidence since as Mandelbrott remarks when there happens phenomena with ten standard deviations it means it needs the time of galaxy to occur :D in practice you don't have to wait so long so the theory must be flawed. BUT EMH is a fuzzy concept, that is it is not true if it is confused with Random Walk as it was the case at the beginning. But since there were explicit evidence they just say well ok Random Walk cannot be BUT EMH can be without Random Walk :D So EMH is now in fact a postulate :p. Moreover there is not a very clear definition of Efficiency and entire book could be written just to try to define it : I bought a doctorate thesis that just talk about that :D before creating my model since I wanted to see the official state of art before reinventing the wheel. In fact I almost have to reinvent the wheel since all these academics stuffs cannot be used for trading :)


>Moreover, mathematical analysis of financial time series shows >that the market is NOT normally distributed, it is more like a >Pareto-Levy distribution.
That's why we can have 10 standard deviation sometime without waiting the end of the galaxy :)


>Unlike normally distributed time series which have computable >moments, the market distribution is NOT well behaved - for >example, the Pareto-Levy distribution has INFINITE variance! In >other words, financial time series are NOT STATIONARY and they >are NOT LINEAR.
I do not really agree with that. In practice there are some moment of stationarity in fact most of the time: it is the period when market is coiling. Every traders know that 80% of the time they do nothing than sleeping in their rocking chair :). When market burst they jump in and mixing the two type of periods give a mixed distribution that is not stationary.

As to be not linear yah sure my model is not linear.

>Consequently, EVERY filter you will ever find in any book on >signal processing, be it a simple moving average or an >advanced Kalman tracking filter, is about as much use as a >chocolate padlock in financial applications! What is not written in >two inch high letters on the first page of these books (but what >is always tacitly assumed), is that the signals are STATIONARY >and hence LINEAR filtering is applicable. This is generally true for >signals in communications equipment but it is NOT true for >signals arising from natural phenomena, especially the markets.
As I said market is stationary by period that's why break methods work so well. As for statistic method you are not obliged to use filter but use robust statistical test not assuming any form of probability law. These are called non parametric tests that do not assume normal law. As for me I have a probability table every day (http://perso.wanadoo.fr/harrytrader/market/dji/today/bornes_stats.gif ) that are calculated without assuming normal law.

>What is required for the markets is NON-LINEAR analysis. This >is, of course, MUCH more difficult both theoretically
Sure :)


>and practically than the simple linear cases, so textbooks don't >bother with non-linear stuff; it is just too difficult.
It is not true there are many researchers in the field. But what is published is often useless in practice of trading for the common trader. Perharps for arbitragists but not for me as a speculator wanting to get a real hedge. There are sometimes interesting things but it is not connected with trading.


>Likewise, vendors of technical analysis software generally use >what they can find in the literature, at most tarting up the >algorithms a little here and there. But these algorithms don't >work!
It is a little pretentious to assume that all vendors only sell things that don't work. Sure some of them are using marketing tricks.

>Gauss proved three hundred years ago that the best estimator >of a random process is its moving average. If the markets are >random, then moving averages should make you money every >time. But they don't. Now you know why."
This is rather in contradiction with what have been said above: it was said that market is not a random process. In fact if it was really a PURE random process it would be easy to win. I won't detail but just understand this phrase of a statistician: randomness and certainty are not antinomy, randomness even give global certainty that is why you can calculate the PI number for example with a random process called the Buffon's experience (Buffon was a french mathematician).
 
To understand what I mean with an image, imagine you mix the ages pyramid of americans with the ages pyramid of pygmes well you will not get a normal law for sure :D

>Unlike normally distributed time series which have computable >moments, the market distribution is NOT well behaved - for >example, the Pareto-Levy distribution has INFINITE variance! In >other words, financial time series are NOT STATIONARY and they >are NOT LINEAR.
I do not really agree with that. In practice there are some moment of stationarity in fact most of the time: it is the period when market is coiling. Every traders know that 80% of the time they do nothing than sleeping in their rocking chair . When market burst they jump in and mixing the two type of periods give a mixed distribution that is not stationary.
 
You must log in or register to reply here.
Back
Top