Way, way off topic: timing

Definitely the wrong board! but…

Just for fun, I just thought I’d mention that my “short term market bottom” detector triggered for the last three days running.
Yesterday’s S&P close 3991.24.
This is not the best signal in the world, but it predicts that the market will be higher in a month with an average S&P index rise about +2.35%.
Average rise after 3 months +3.28%, though wider dispersion: this signal wears off quickly, so I usually say it’s good for 1-4 weeks.

This is the 64th such signal in the last 56 years, counting consecutive days as one distinct signal.

This is just a so-so quality predictor of a short term bounce.
No attempt to call a major market bottom.

For the record, Berkshire Hathaway B closed at $312.96. Last trade at $313.20 in pre-market as I type.
Though this is a (bad) prediction about the broad market rather than any single stock, Berkshire’s price correlates pretty closely on short time frames.

Jim

32 Likes

Just for fun, I just thought I’d mention that my “short term market bottom” detector triggered for the last three days running.
Yesterday’s S&P close 3991.24.
This is not the best signal in the world, but it predicts that the market will be higher in a month with an average S&P index rise about +2.35%.
Average rise after 3 months +3.28%, though wider dispersion: this signal wears off quickly, so I usually say it’s good for 1-4 weeks.

This is the 64th such signal in the last 56 years, counting consecutive days as one distinct signal.

This is just a so-so quality predictor of a short term bounce.
No attempt to call a major market bottom.

Just for fun, let’s play devil’s advocate.

The S&P has risen about .7% per month over that 56 years anyway, so you’re only gaining about 1.7% over that at best. And that’s actually attained over a year, if the signal fires once per year or so.

Also we don’t know what the dispersion was, which is necessary to know how much we should bet.

Also this is presumably mostly/all the result of backtesting, and we don’t know how many different signals you tried to fit to historical data before finding this “so-so” one. This alone is probably enough to avoid using it.

Also, gazillions of dollars go into using machine learning to find useful short-term signals, and they have access to the same historical data, so the signal has probably been competed away by now anyway.

And even if the signal did work as advertised, you could only take advantage of it by holding some extra cash for the occasions when it fires. Since it fires about once per year, you would hold that extra cash for 6 months on average, during which you would lose more on average from not having it invested, than you would gain back with the timed buy.

11 Likes

The S&P has risen about .7% per month over that 56 years anyway, so you’re only gaining about 1.7% over that at best. And that’s actually attained over a year, if the signal fires once per year or so.

Also we don’t know what the dispersion was, which is necessary to know how much we should bet.

Also this is presumably mostly/all the result of backtesting, and we don’t know how many different signals you tried to fit to historical data before finding this “so-so” one. This alone is probably enough to avoid using it.

Also, gazillions of dollars go into using machine learning to find useful short-term signals, and they have access to the same historical data, so the signal has probably been competed away by now anyway.

All reasonable comments.
Market timing is not going to make you rich.

I think of the signals much as I think about weather reports.
They are highly fallible and you would never bet the rent on them.
But they are still useful, statistically.
Some days you head out with an umbrella, some days you grab the picnic blanket.
Some days are good days to lighten up on things you were likely to lighten up on anyway.
Some days are more appropriate for deploying capital on things you like anyway.
If you’re going to sell your XYZ stock some time soon, why not do it on a statistically auspicious day?

A thing does not have to be 100% reliable to offer a bit of an edge.
What are the odds in favour of card counters? Miniscule.
But a bit of an edge is better than no edge, if you care to use it.

As for the models having been tuned on past data…well, sure. But that was about 15 years ago.
The same basic model (more stringent cutoffs) let me nail both the 2009 and 2020 market bottoms within about 3 days, and others, publicly posted at the time.
Whether and how one uses the information is a matter of taste.
Like a weather report.

Jim

40 Likes