I have noticed as back as in 1995 and in the case of some commodity futures that you could even predict some series with 80% probability of success. Still, what these authors are missing is the fact that winners and losers tend to average to 50% minus commissions, meaning that such naive approaches cannot provide any edge, or put in other words, the series itself cannot be the edge. In practice you need to exit at some place. As soon as you introduce an exit, you introduce losers and that is the end of the game as far as this approach. I remember I was so frustrated when I discovered I could predict next day's direction (up or down) with 80% probability but still I could not make money in the longer term. The 20% that I missed contributed to as much losses as winnings. This is the nature of the markets. To make money you must do better than naive statisticians.
An amazing thread. All that is being spoken about are "traps". The meaning of the word trap is: by accepting huge arbitrary givens, the meat of the opportunity is foregone. Formal education involves an expenditure of time. The setting is coherent for the most part. Making money is the thread's presumed result (Profitable System). Is it not possible to consider Carnap and Keynes and dump the Bayes and "frequentisits"? Isn't the market providing information? Can't this information be classed as non probabilistic? The fact is: is that ALL information from the market is real and in an order.
So what? Another fact is that I just ate a couple of delicious enchiladas. The thread has a specific topic. It isn't "is looking at derived data worthwhile?" Your post is (as usual) completely worthless.
My posts have little value to most people. Check the title to find where I got the words "Profitable System" in my post. "Profitable System" is a reference to an expected result. You feel the topic of this thread is not "is looking at derived data". Any process whereby data is processed in any way is often called "derivative data". Quant machinations on raw data inputs from markets is generally considered "derived data". My point was that people are using end of bar data as a "given" to facilitate whatever they are doing. When precision is a requirement, end of bar data is NOT used. Using any kind of bars is also an initial crude decision to facilitate whatever is being done. Several contributors spoke of things they have done aftr the two above "facilitators" have been put in place. As you see, often these activities are of less import than the two conveniences originally put in place. In college or university. this sort of errror is usually not afforded in the formal learning process. University graduates who are participating here know that and they also know that the labor they put into getting eductaed is sufficient to allow them to not hunt in unfertile places. I am glad the topic here, as you put it (for yourself) isn't for deciding whether looking at derived data is worthwhile. The OP has already pointed out that he is just doing an exercise to further his prowess in the application of appropriate knowledge. He knows he will not be dealing with a "Profitable Sysytem" simply because he is so far off the reservation. Any profitable system development begins with assuring that ends of bars and a specific time frame is off the table. As I said, most of my posts are useless to most readers. There are a lot of reasons why this is true for these people. Try a Sonoran dog sometime.
Then why do you continue to post? Just like to see your name in colored pixels? The fact that your posts are 99.999% nonsensical drivel doesn't make you interesting or insightful. You write like the idiot PhD candidates you like to mock. If you have something to say, just say it. If you can't say it in a way that is meaningful to your audience, figure out a different way to stroke your epeen. You like to say crap like "quant community", as if you're a participant, an expert, or a contributor. Seems to me there are two ways to gain credibility: obviously know what you're talking about so no one needs further evidence, or point to things you've accomplished, so that people have some sort of concrete reference. An English physicist (I think, and can't remember his name) said something along the lines of, "If you can't explain your work to an English barmaid, you don't know what you're talking about." Feel free to respond, but I won't return the favor. Hopefully this thread doesn't have to become one of the hundreds (I found after about two minutes of looking) derailed by your participation.
Yeah, there's at least one error in that paper if memory serves. I'm sure you know that E(X^2) is not 1/2 if X is a uniformly distributed RV in [0,1], so the correl in the first example is not -1/4 it is a bit lower (well, greater but they are talking about anti-correl) at -1/12. I think Sornette can come across as crackpot, but once in a while you can grab some ideas from his ramblings.
SomeYoungGuy, I looked through my bookmarks and did not find anything similar / thought-provoking. But I guess Wikipedia articles on the same topic(s) should be useful.
I took a look at that paper and I could be wrong in my analysis, but the results seem trivial. The E[X|x>.5]= .75 assuming a Continuous uniform dist. from 0 to 1. So if we are given a value of x, s.t. X>.5 or x<.5. Then our cond. expected values will be .75 and .25 respectively. So how often will any x be less than .75? of course it's 75% of the time. Does this seem like prediction? If I referenced any value in the sequence that was < or > .5 I could say that 75% of the time I can predict the increment. Regarding the synthetic data I say start with a simple random walk and start modelling and developing and then work up different stylized facts in order to isolate the relevant features.
Hi Mike, Thanks for the thought experiment. When I first glimpsed at this thread I thought you were going somewhere else with the experiment before I had a chance to look at the spreadsheet. Was the motivation for this experiment optimal stopping or was is bayes Monty Hall stuff? Have you checked out the bandit problems? From what I can remember there's a couple guys that have done a lot of research on that types of stopping problems that have tie in with Jim Simons. Samuelson has an interesting rebuke of the the kelly criterion using the optimal stopping problem and martingales as his motivation. pt
I don't have time to catch up on the details in this thread now, but I think I have been down this road before. If this is what I think it is, the trick is that this type of algo works on a stationary price series. At best stocks will have an approximately stationary return series, but not a stationary price series (unless maybe you are very high frequency). But you can create a portfolio with an approximately stationary price series - long and short a pair of cointegrated stocks.