|
PureBytes Links
Trading Reference Links
|
At 1:24 PM -0400 10/13/98, Neil Harrington wrote:
>I would like to hear from the list about any of the following thoughts:
The term was coined by Bob Brickey. I quote from his post on the TradeLab list:
>A truer view is that most trading indicators, whether of the mathematical
>or human judgment variety, range in value from random to a blend of random
>and true. Most have reliabilities no better than random chance. The best
>indicators ever devised provide a blend of true and chance indications.
>The objective is to find indicators that provide the best possible ratios
>of true to random chance indications.
>Some functions can be correctly classified as "gambler functions," because
>it can be shown by means of theoretical analysis or practical tests over
>large numbers of trials that they provide no advantage compared to a
>random selection process.
<End of quote>
Back to your questions.
>1. Many on this list readily talk about gambler indicators not working,
>mainly because they are all lagging indicators, telling you (as Allan just
>said, "They tell you where things have BEEN", not necessarily where they
>are headed.
As it turned out many people, including myself, have discovered that they
are of limited use in trading systems since they do tend to lag the price
action and give you lots of spurious signals.
>2. From discussions on this list, gambler indicators are apparently any
>and every indicator you read about in all of the books on technical
>analysis.
Most of the indicators were designed to point out some characteristic and
they do a pretty good job of that when used as an INDICATOR. This is
because you are looking at historical data. In such cases, you will find,
for example, that MACD or Stochastics always tends turn about when the
price is turning. I sometime use them to search through historical data to
try to find some characteristic of the price that might be tradable. They
are very similar, in concept, to the "SwingHighBar" and "SwingLowBar"
functions in TradeStation. These are PERFECT at pointing out all of the
tops and bottoms of the price curve, but unfortunately 5 or 10 bars after
the fact.
>3. If #2 is false, please list some popular non-gambler indicators.
A mentioned above, the functions are pretty good as INDICATORS to help you
spot some characteristic. The problem for trading systems is that they
usually look good as indicators because they contain various averaging
(smoothing) functions which cause lag, by definition. And lag gives you the
buy/sell signal in a system too late in many cases.
>4. If #2 is true, please list some non-gambler indicators that work, or as
>Allan said, "Is there information in the price series and other
>intermarket information that can give short-term predictability?" (I
>thought this was the objective of all indicator development).
A trading system needs to capitalize on some characteristic of the price
that is at least somewhat predictable from what has happened in the past.
You need to find such a characteristic to create a tradable system.
For example, it is pretty apparent that the S&P 500 index tends to go up
for a few days then down a few days. A very simple system can take
advantage of this:
if Close > Close[1] then Buy next bar at market;
if Close < Close[1] then Sell next bar at market;
This was mildly profitable (43%) with a Return on Account of 342% over the
past four years.
But since you know the trend has been mostly up over this period, you might
expect that the following would be better:
if Close > Close[1] then Buy next bar at market;
if Close < Close[1] then ExitLong next bar at market;
This was more profitable (51%) with a Return on Account(ROA) of 627% over
ten years. Not too shabby. And if you look at the equity curve, it builds
up very smoothly, resulting in a pretty high Sharpe Ratio. It is a much
better system than most of the canned systems that come with TradeStation
(which use the popular functions).
Does this use an indicator? No. It just uses a characteristic of the price
that I happened to notice.
Is it "rocket science"? Hardly.
Does it make logical sense? Obviously.
Is it easy to understand? Obviously.
Now tell me if you really understand what the following buy condition will do?
if ADX > 40 and MACD(Close,9,19) > 10 then Buy;
I don't.
>5. Gambler indicators are all basically logically-based and easy to
>calculate. The impression I have of non-gambler indicators is that they
>are rocket science, not published, not generally available to the public,
>and thus the private domain of a few rocket scientists. I consider myself
>"smarter than your average bear", but not a rocket scientist, and
>currently know of no indicators that I think the list would consider
>non-gambler indicators.
As noted above, it doesn't have to be rocket science. It just has to take
advantage of some characteristics that are reasonably predictable. I think
you would find that most of the better systems being used are pretty
simple. I have one that uses only four lines of TradeStation code that has
an ROA of over 1000%, annualize return of 52% and a Sharpe Ratio of 2.44.
It takes advantage of a simple price characteristic I noticed. (Obviously,
if I tell people what it is, it will stop working.)
To find a tradable characteristic, some of the standard statistical test
are useful to confirm that the characteristic you think you see is
statistically valid. I use filters, such as Tillson's T3 filter or Jurik's
AMA filter, to try to separate the characteristic I want to trade from
everything else. These are about as close to rocket science as you need to
get.
>6. The comments on the list about this topic unanimously talk about the
>uselessness of gambler indicators and the superiority of non-gambler
>indicators, but then nothing is ever offered up about specific non-gambler
>indicators that work and how they work. Yet specific talk is always
>available when someone has a question about EL code for this or that
>variation of a well-known gambler indicator, as if everyone is really
>using gambler indicators in their work. Why is this?
I think the reason is that they aren't indicators. They consist of code to
take advantage of some characteristic that tends to repeat as in the
example above. I could make the condition of my trading system above into
an indicator:
if Close crosses over Close[1] then Plot1(Low, "Secret.Buy");
protect it with a password and sell it. But Mark B would figure it out and
beat me up publicly. :-)
>7. What about improved gambler indicators, like Mark Jurik's JMA, the
>Visine of MAs that "gets the lag out"? Does this make it a non-gambler
>indicator?
I wouldn't call the Jurik AMA an "improved gambler indicator". It is simply
a filter that tends to track price closely while removing noise. It can be
a useful component of a system if your trading idea requires this sort of
filter function.
>Thanks for any thoughts.
I know it must seem confusing. I sometime think the infrastructure of the
industry makes it seem complicated so that they can sell us lots of books,
indicators, trading systems, and complicated software packages. We are told
that there are Trending Systems, Support and Resistance Systems, and
Volatility Breakout Systems. Unfortunately, the markets I trade don't know
which of these they are suppose to be at any given time. We are told that
the market is "oversold", but it doesn't understand this and keeps going
down.
Eventually, after you get tired of "The Indicator of the Month Club" you
figure out that the price just goes up and down. And, as somebody real
smart once said, the trick is simply to "buy low and sell high". Everything
else is superfluous.
Bob Fulks
|