Richard, I will have to disagree with a few points that you have made.
Futures like stocks have subsets which respond at different speeds. Some move fast like internet stocks (look at a chart of silver) some move slow like cyclicals. What an individual chooses to trade depends on their tolerance for risk versus reward.
But, there is nothing inherently slow about futures.
Futures of an index like S&P 500 is simply the cash value of the index plus the daily carrying cost until the expiration day.
If the Cash or Future price become divergent from the other, arbs in the market will close the gap.
It is true that some systems work better on Stocks. While other systems work better on Futures. Some systems will work on both Stocks and Futures.
I will post a couple paragraphs from the previously mentioned article. Since i have personally tested the systems mentioned and I know that they are some of the very best systems available. These are the same systems that I personally trade on a daily basis.
Ruggiero: But, I mean, if the system doesn't work in the future, the risk of course is unlimited. That's why I use a simple
measure. I'll optimize a system over an entire grid of values, take the average, standard deviation, skew and kurtosis and do
some numbers on that, just to see if I have something stable. If the average across the board of the optimization is not
greater than the standard deviation, then you have nothing. What you find out is for things like moving average crossover or
dual moving average crossover, or any indicator-based system, the only classic system that at least has a one-to-one ratio is
channel breakouts. And if we look at real life, most of the CTAs who are making money are using some version of
breakouts, which has held up. So, the statistics show that. So we need to add better measures that answer, What's the
probability that a trader or a system will work in the future? What's the probability on future drawdown? That is, if you
assume the method works, is it going to blow up anyway? And what are the flaws in his method? We have to go back and
understand what he's doing, understanding the premise and rationalizing a theory about the premise to define risk. No, but
you might be able to define models. Let's say Long-Term Capital Management, if you look at non-normal distributions and
say, "Wait a minute, five sigmas are going to occur with this probability." You could say, "Guess what. There's this
probability they're going to go under."
FM: What are those measures that we need to have in the software?
Knight: From my standpoint? For example, just some very simple statistical stuff the student's t-test, which is very imperfect
as we alluded to earlier, on the trades, [which answers] what's the probability that this is going to be a winning system in the
future. The t-test gives you an answer to that. And it pools the percentage winning trades, the profit-and-loss ratio, all this
stuff is brought together into a single number. And you can say based on the student's t-test, with all of its imperfections,
there's a 90% probability that this system is going to make money in the future - or a 20% probability. Now, you have to
decide what probability or what odds you are willing to take. If you're very risk averse, maybe you got to have 95%. Well,
you're not going to find many systems. If you're less risk averse, maybe you'll settle for something like 70% or 80%, but
that's an example of a single number that can be gleaned from all the stuff that comes out that's not currently available, unless
you want to program it yourself and put it in as a function, as you can do, but it's not available in the standard packages.
And there are other things, too. There's a whole other field called non-parametric statistics, which work a lot better when
you're dealing with non-normal situations. There's a non-parametric regression technique that I've played around with a little
bit that works much better than the standard multiple linear regression method that people use.
FM: These non-parametric methods are the ones, too, that are more available now that we have more computer
power.
Knight: That's right, because they are much more computationally intensive than the standard methods.
FM: The reason we have statistics is so we can assume things like normal distributions and then infer
conclusions based on the idea that the data falls into this distribution...
Knight: Non-parametric techniques are much less sensitive to the assumption of a normal distribution. They handle outliers
much better. You can even have five sigma cases out there and it doesn't distort the results much.
Ruggiero: Actually, Sheldon has made an interesting point [before], which is if you add a volatility breakout on top of
almost any trend-following method to filter out noise - because here's the thing, with any breakout technique - and, Sheldon,
you can say this - it is the question of noise vs. signals.
Ruggiero: Actually I proved - I took channel breakouts and put opening range breakouts on top of it and proved that that
it outperformed standard channel breakouts.
Ruggiero: But I will tell you something real quick, just dealing with this astrology. Let's look at the bond market for a
second. Bonds and commodities. I believe that astrological stuff is somewhat valid in those markets for a couple of reasons.
I used to be involved - my undergraduate degree was in physics astronomy. I did a lot of astrophysics work. My specialty
was long-range weather forecasting using astrophysics. And I did a lot of harmonic models of sunspots based on the
harmonic positions of the planets. And our sunspot cycles and the cycles shifting are controlled by longer-term planetary
stuff. And because I have sun spot data going back to 1720, you can see this stuff and actually be valid because you have
250 years of data. If you look at that stuff, if you believe that sun spots will affect weather patterns, so commodity prices go
up, you can say that bonds are going to have problems if commodities go up. So you can come up with a link between
astrology and the bond market or commodity prices.
Bianco: Paul Montgomery does a lot of astrological work, and his argument is even more basic than that, that because the
human brain operates on electrochemical impulses, you know, electricity jumping the synapses. And all of these astrological
events, well, some of them, do affect that behavior...
Freeburg: A great example of that. Going way back to the beginning of the whole debate was a guy named Sidney
Alexander, who was a professor of finance at MIT, and he claimed he could use what Marty Zweig later called a percent
swing system where something goes down - say the Value Line index or T-bonds - goes down, hits the bottom and rises by
a percentage - let's say 1% - then you buy. It reaches, then it drops by that 1% and you sell. And this reciprocates back
and forth. He claimed that a strategy like that, a simple trend-following strategy, could outperform the market, and a young
recent graduate challenged that. This was back in the '60s. He was the leading academic theorist. He became one of the
principal architects of the efficient market hypothesis, kind of our intellectual adversary. Well, now it turns out with
diminishing slippage and commissions, you can use strategies like that to trade the Nasdaq 100 on a very small percent
swing basis and it will work.
Ruggiero: Like on the S&P, we're going to test yesterday's close with a 90% probability even though we opened 16% of
yesterday's range above that close. Well, the only problem is occasionally the market just blows up in your face and takes
off, but if you're not stupid and you're looking at tick and trin and stuff so it doesn't happen, these little trades - 15% of the
range is only a point and a half - so the disaster trade would blow you out of the water. But with no slippage and
commission, being right 90% of the time, those trades are workable.
Knight: A lot of the trading techniques that we use now actually have done that. For example, one of the principals of
human behavior is people tend to overreact. And all of these gap trading systems, which are very effective and have been
for a long time, all of those systems are basically a way of trading off the fundamental fact of human behavior that people
overreact in price. |
