|
PureBytes Links
Trading Reference Links
|
The problem with velocity and acceleration is that they are
derivatives (velocity = 1st derivative, acceleration = 2nd).
And the process of calculating derivatives AMPLIFIES NOISE
in the underlying signal and introduces its own souces of
noise.
Commodity prices are plenty noisy already; computing
derivatives (called "numerical differentiation") adds more.
What you wind up with, is noise noise and more noise,
and very little signal.
An appealing-sounding twist is to work, not with raw prices
directly, but with "smoothed" data series ... generally, lowpass
filtered. The idea is, the derivatives of "smoothed" data will
be less noisy. This amounts to integrating price (since lowpass
filters are integrators), then differentiating it to get
"velocity of smoothed price", then pretending that something
useful results from the exercise. At least in my experiments,
it hasn't.
Rather than looking at "accuracy percentage" or "correlation
coefficient" or "root mean squared error", I look at
"net profitability of a mechanical trading system including
commission and slippage." Perhaps this criterion is too
harsh; I guess it's a matter of personal taste. For those
who disagree and who desire a method that's >80% accurate,
here's one for free. But its net profitability ain't so good.
If position = flat and today = monday then
buy at market on close.
When filled, place a limit order to exit long
at (EntryPrice + 3 ticks), good till cancelled.
Place GTC StopLoss order to limit loss to $20,000
per contract. E.G. in ten year treasury notes,
StopLoss price is EntryPrice minus 20.000 big points.
Best regards,
Mark Johnson Silicon Valley, California mark@xxxxxxxxxxxx
"... The world will little note, nor long remember, what is said
here today..." -Abraham Lincoln, "The Gettysburg Address"
|