|
PureBytes Links
Trading Reference Links
|
At 9:37 AM -0600 7/7/98, Gary Fritz wrote:
>Bob Fulks wrote:
>> I don't know how to explain it in a short message without pictures
>> but will try.
>
>Bob, thank you for an *outstanding* post. That's a wonderful
>description of how to apply a logical and scientific approach to
>system development, instead of the "throw some indicators at the
>wall and see if they stick" approach. If I could, I'd like to ask for a
>bit of expansion and clarification of some of your points:
>
>> characteristic in real time. In this case we are looking for a periodic
>> component that occurs around every 10 days. So we would like to isolate
>> this component from the other information in the price pattern to see if we
>> can generate trading signals from it. We design a filter that passes data
>> with a period in the range of perhaps 5 to 15 bars. I usually use
>> combinations of various moving averages for this. (A moving average is a
>> filter that can be used to remove periodic components with fewer than some
>> number of bars.)
>
>Can you expand on this a bit? My EE filter-design classes were 20
>years ago, and I was never very good at it. :-)
>
>I assume an N-period MA (do you usually use an XMA, or maybe
>Tim Tillson's T3 because of its faster response, or...?) acts as a
>high-pass filter to remove signals "faster" than some period X, where
>X is probably less than N, but some specifics would be very helpful.
>E.g. if you have a rough rule-of-thumb relating N and X, that would
>be great. For now I just use the "plot it and see how it looks"
>method.
Since you are an EE (as am I), I will use that terminology. I use a lot of
different filters but the T3 filter is fine for this purpose. The numbers
in the post are a good start (5 to 15 bars period). That would be 17 to 45
cycles per year on daily data. You need to watch the phase shift since this
causes delays.
>
>> You can then create an indicator to plot the resulting filtered signal
>> to compare it with the original price signal. It should be a lot
>> smoother and have the trend removed so that it oscillates around
>> the zero line. Many common oscillators such as MACD have
>> a similar structure and use similar moving average filters.
>
>So do you use something like MACD's MA(fast)-MA(slow) to remove
>the trend? Presumably with the slow MA set to a period long enough
>that it removes the oscillations of the signal you're interested in,
>i.e. so MA(fast) contains the signal but MA(slow) doesn't?
The high-pass filter will remove the trend (DC component) and the low-pass
filter will remove the high-frequency noise (to the extent of the cut-off
rate of the filters). I do not use MACD. I was referring to the fact that
MACD also uses filters in a similar way. Sorry it wasn't clear.
>
>E.g. as a first experiment I tried plotting
> Xaverage(High,5) -XAverage(High,15) on a daily SPX chart,
>and it does a reasonably good job of isolating the tops -- and the
>bottoms too, since the swing bottoms tend to have low highs.
This is fine but the XAverage filter has a cutoff rate of only 20 db per
decade which is pretty slow. T3 may be better. Avoid using filters with
lots of poles and zeros - they have weird phase behavior.
>
>> Ideally, the tops and bottoms of our resulting
>> signal will occur at or near the same bars as the tops and bottoms of the
>> ShowMe we created above. We can test this with correlation functions and
>> adjust the characteristic of the filters so that we get good correlation of
>> tops and bottoms.
>
>Can you explain exactly what you mean by correlation functions?
You can run correlations between the price series and the trading signal to
maximize the correlation as you adjust the parameters of the filters. The
objective is to have the two line up in time.
>
>> Now we need to generate signals near the tops
>> and bottoms of our trading signal.
>> Simple things such as:
>> if Sig > Sig[1] then buy
>> if Sig < Sig[1] then sell
>> work pretty well but lag a little. Let's use
>> this in our example.
>
>The lag is a problem with many indicators, and turns many good
>indicators into money-losers. Do you know of a good (safe,
>reliable) way to anticipate the turn in the indicator and get aboard
>the move at the start?
The above would trade the first bar that the signal changed direction.
Depending on the entry technique, MOC, market order, etc., you might want
this for entries - a little later but surer. An exit before the top of the
signal might be better to avoid giving back some of the gains.
>
>> Now we probably get too many unprofitable trades. For example, we probably
>> want to trade only with the trend (when the trend is up, we want to go long
>> and flat, when the trend is down, we want to go short and flat, and when
>> the trend is flat, we may want to go both long and short). So we need a way
>> to determine the trend. (There are lots of ways to do this, none perfect.)
>> So we then use a trend signal to filter out trades. Assuming a Trend
>> function that returns values: up = 1, down = -1, and no trend = 0
>
>I want that Trend function!! :-) I've been trying to find a good
>indicator of trend, *especially* one that detects a no-trend condition.
>I (and probably most other system developers) could vastly increase
>the profitability of my systems if I could figure out how to keep them
>out of the market in flat periods -- or, better yet, switch to a range-
>trading strategy when the market flattens.
>
>Obviously the Trend function would need to be tunable for different
>measures of trend (e.g. are you interested in the 10-day SPX
>period or a longer 2-month super-period?), but that should be
>do-able.
>
>I'm sure none are perfect, but I'd appreciate some pointers if you
>know of some good trend indicators.
>
I've tried multiple length moving averages (price above all is uptrend,
etc.), consecutively higher swing high bars, etc., and about everything
else I can think of. As I said, none are perfect. I keep looking.
>Again, thanks for an excellent post!
Glad you liked it.
Bob Fulks
|