"The hard part in my mind is working out what the trend is half the time "
That is the relatively easy part. The hard part is what do you do with the information.
For food for thought here is a bit of research by Victor Neiderhoffer.
"Continuing on our eternal quest to raise the level of logic and science in the field of investment, we did the unthinkable in our field. We actually tested “The trend is your friend,” which is, of course, the main shibboleth of technical analysis. The test is very simple, the kind of thing that any high-school science student would be required to complete for a passing grade in a weekend homework assignment: Is there a positive or negative correlation between consecutive changes over periods of the length most people mean when they’re talking about a trend?
The actual length of the period is not terribly important, as almost all typical measures of trends within time periods of, say, 20 or 30 days, are quite close to each other. If one measure, say, the 25-day moving average, is up, then the 25-day regression or 30-day line between consecutive lows or what have you is also going to be up. We chose as our measure of trend the move in the last 20 days in S&P 500 futures. To measure the trend’s tendency to persist, we calculated the correlation between the change in the last 20 days and the change in the next 20 days. That correlation is -0.12 over the last eight years
The numbers indicate that when stocks have gone up over 20 days, they are likely to go down over the next 20 days, and when they have gone down over 20 days, they are likely to go up over the next 20 days.
Recall here that the correlation measure is the linear relation between two variables, with 0 showing no relation, 1 a perfect positive relation and -1 a perfect negative relation. The actual level of -0.12 is quite high for correlations involving stock market predictions and indicates a strong tendency to reversal. Thus, our rule to find out what the trend followers are doing and to do the opposite is a good one.
While a negative 12% correlation might not be considered overly large in magnitude, little things have a way of adding up in the stock market. The average S&P 500 futures move in the 20 days after all those occasions when the stock market was down in the previous 20 days is up 0.8%, or about 9 points. The 20-day periods after a rise in the previous 20 days showed almost no change, on average. (For the purists, the average change was -.01%.)
A 6,000-point edge When you consider that the S&P 500, adjusted to today’s levels, rose some 500 points during the eight-year study period, the 6,000 extra points that an investor would have realized from buying on the 722 occasions when the 20-day average was down can amount to a significant figure. Six thousand points is quite an edge to have.
We doubt that a simple demonstration such as this will be accepted without many objections and diatribes. Something as simple as a test of its validity would naturally be resisted as possibly leading to a new paradigm. While we don’t consider ourselves heroes of the Galileo mold in showing that gravity affects all objects equally, we do consider the state of knowledge in the field of technical analysis much less advanced than that in any of the sciences in, let’s say, the 15th century.
One critique we are sure to hear is that our test is unfair in that it did not use 25-day or 30-day trends. The complaint would have been valid had we not also computed the correlations for longer periods and found that the correlations are even more negative.
It will also be objected that a simple linear measure of association like correlation is not adequate to capture what's really going on in the stochastics, waves, tests, heads and shoulders, and points and figures of the market. But no matter how you cut it, the overwhelming negative correlation would make any effort to overcome it with an ad hoc postmortem totally ineffectual."
