Trend Following Research

[The Big D]

Quote from marketsurfer:

Take a randomly generated chart. Make a good guess, go long right before a 20% run up. Guess right again and take your profits. Then the random up move drops, you don't trade again thereby beating the inherent rate of return generated by the random (Quasi) chart.

In other words, luck will cause some traders to be successful regardless of randomness or lack there of.

Hence why confidence intervals are used to separate true non-random effects from the random oddities of a small sample. That's why for my example I offered up a method that has 1500 samples and a confidence incredibly close to 1.

In other words, what you've got there is a weak argument.

 

[Gabfly1]

Quote from marketsurfer:

...In other words, luck will cause some traders to be successful regardless of randomness or lack there of.

And that's what you're pinning your hopes on?

Incidentally, you did not respond to my previous post yet.

 

[Neenisti]

Quote from Gabfly1:

And that's what you're pinning your hopes on?

Incidentally, you did not respond to be previous post yet.

How about all of the posts the OP hasn't responded to?

 

[Gabfly1]

Quote from Neenisti:

How about all of the posts the OP hasn't responded to?

Actually, it was Mike Covel who started this thread. As for marketsurfer, Perhaps he has concluded he does not have an edge in this debate and is cutting his losses:

Quote from marketsurfer:

...Exactly, without ironclad risk control and position limits it can be disastrous.

http://www.elitetrader.com/vb/showthread.php?s=&postid=3155705#post3155705

 

[Neenisti]

Quote from Gabfly1:

Actually, it was Mike Covel who started this thread.

That's who I was talking about.

 

[marketsurfer]

Quote from Gabfly1:

But against the very background you described, and as a self-proclaimed "quasi-random walker," are they in fact "charts of randomness," or are they at times the depiction of "edges" playing themselves out over time. Not instantaneous edges, but arguably ongoing ones that an attentive trader can latch onto. If you have special information and somehow manage to catch a move at its very onset (good luck with that), would there not be time for someone less plugged in to see a bit of directional price action taking form, the very price action that the very informed you are presently exploiting, and hitch a ride? Do you think that once the early riders are on the edge bus that no other passengers can come on?

Youy really don't take the time to think your own theories through to their logical conclusion, do you?

Are you stating patterns repeat in an exploitable manner? Fortunate for the markets existence, they do not. Sure folks can jump on a move but the move is only recognizable after it happens and it can alter at anytime.

 

[marketsurfer]

Quote from Gabfly1:

And that's what you're pinning your hopes on?

Incidentally, you did not respond to my previous post yet.

Van Tharp claims to have shown that random entries with good money management can be profitable. "I would rather be lucky than good "

Instead Of just believing what "winner take all" erroneously states about Williams Larry you should listen to the rest of the book. You could learn something.

 

[marketsurfer]

Quote from The Big D:

This whole discussion seems to be wanting for an example of a working trend following system, so here's one:

On weekly bars the S&P has an upward bias in excess of basic returns when a given bar opens over the 40 bar EMA. It has a negative bias when a given bar opens below the 40EMA.

The statement I just made is easily verified, it's tradeable, it's highly statistically significant, and it's been that way for decades. It's classic trend following.

So to get down to brass tacks, in certain instruments on certain time frames there's ironclad evidence of trending even using a trivial detector for trends. Now discussing how best to find the right instruments and timeframes and how to best follow the trends thus discovered is a broad topic. But let's not pretend there are no statistically significant trending instruments out there.

I consulted with one of the smartest market math guys I am friendly with regarding your claims. Although, I know basic stats, I am far from an expert. He advised that over the last 105 days, the correlation of daily SPY with itself the previous day has been negative 10%. The market is negatively auto correlated at lag 1. So whoever made the statement that it only takes one up day to increase odds of success is wrong.

However, MA crossovers can have some significance at the 40 week level and actually begin to work at a little more than 100 days.

I stand corrected with my previous statement.

Surf.

 

[Neenisti]

Quote from marketsurfer:

So whoever made the statement that it only takes one up day to increase odds of success is wrong.
Surf.

Your original statement was "How many moves in one direction increase the odds that the next move or series will be in the same direction?"

One day is not a single move. One up day is meaningless. It's ignorance just like saying one tick constitutes a trend. One day does not make a trend. There are many price swings from resistance to support each day.

To put it in context of your initial question, How many "price swings" in one direction increase the odds that the next "price swing" or series of "price swings" will be in the same direction?"

My answer will still be the same, one. I can prove mathematically I am correct. Your math guy is correct based on what he is looking at but his dataset is wrong. I'm currently busy on a project but give me about 30/60 days and I will show you and your math wiz you are both wrong.

You said yourself, you don't understand it so don't try to wish it out of existence or try to explain it away.

 

[The Big D]

Quote from marketsurfer:

I consulted with one of the smartest market math guys I am friendly with regarding your claims. Although, I know basic stats, I am far from an expert. He advised that over the last 105 days, the correlation of daily SPY with itself the previous day has been negative 10%. The market is negatively auto correlated at lag 1. So whoever made the statement that it only takes one up day to increase odds of success is wrong.

However, MA crossovers can have some significance at the 40 week level and actually begin to work at a little more than 100 days.

I stand corrected with my previous statement.

Surf.

I think we're in general agreement on these points - what your friend told you about the day N vs. N+1 correlation is what I alluded to when I said that many markets are inherently counter-trending on shorter time frames and trending on longer frames. Generally speaking the best detectors (MA crossover, serial correlation, etc) for one vs. the other are not exactly symmetrical either.

Of course, the fact that many markets have positive serial correlation on long time frames and negative correlation on short time frames should suggest a trading system concept with two re-enforcing entry signals. And I know from experience that that class of systems can work very well, although the S&P is not the optimal instrument.

My real point was to demonstrate that there is at least one trend following system that works well. And further, that it detects trends which are not only observed facts in the past, but also predictive of the future.