The Quantitative Review
By Andrew Garcia
A daily review of trading rules, tested against chance
An archived edition, as published that night. Read today’s edition
Long Treasuries rose 1.9% at the last close, about 3.0 times their usual daily move. 30-day momentum bought at the close.
Who “the rules” are: 100 computer programs, each following one textbook trading recipe (ten recipes, such as “buy when the 50-day average price rises above the 200-day,” each run on ten markets). No people and no money are involved; the paper runs them on real prices to see whether any beats luck.
Each square is one strategy on one market. Select it to see its year, its trades and its verdict.
Blue: in the market and gaining since it bought. Red: in and losing. Dashed: out of the market. Tap a square for its numbers and verdict.
| Market | Golden cross | Cross 10/50 | Breakout 20 | Turtle 55 | RSI 30/70 | Bollinger | MACD | Buy the dip | Momentum 30 | Momentum 90 |
|---|---|---|---|---|---|---|---|---|---|---|
| Crypto9 of 30 in. 25 bps a trade. Last close Oct 22. | ||||||||||
| BTC3 of 10 in | ||||||||||
| ETH3 of 10 in | ||||||||||
| LTC3 of 10 in | ||||||||||
| Stocks and bonds11 of 30 in. 5 bps a trade. Last close Oct 22. | ||||||||||
| S&P 5005 of 10 in | ||||||||||
| Emerging markets2 of 10 in | ||||||||||
| Long Treasuries4 of 10 in | ||||||||||
| Commodities10 of 30 in. 5 bps a trade. Last close Oct 22. | ||||||||||
| Gold1 of 10 in | ||||||||||
| Oil7 of 10 in | ||||||||||
| Copper2 of 10 in | ||||||||||
| Currency4 of 10 in. 5 bps a trade. Last close Oct 21. | ||||||||||
| Dollar vs sol4 of 10 in | ||||||||||
Markets
Long Treasuries rose 1.9% and litecoin fell 4.4% at the last close.
The least usual move was in long Treasuries: +1.9%, about 3.0 times their typical daily move, the kind of day they have only a few times a year.
Over five sessions, oil has done best, at +3.3%, and gold worst, at −3.6%.
5 rules changed their mind at the close.
Strategy desk
It beat 3 of the 5 coin-flip rules run on the same market: random buy-and-sell decisions, which are the bar for luck.
After charging it for all 150 tries, its chance of being more than luck is 2%, short of the 95% this paper requires. Put another way: if you tried 150 random rules, you would see a run like this almost every time.
The closest any rule has come is 20-day breakout on ether, at 87%.
At the other end, Turtle breakout on copper is down 16.6% over the same 90 days.
If bitcoin went up this week, is it more likely to go up next week, or to give it back? That is all this section asks, for four lengths of time: a day, a week, a month, a quarter. For each, we take every stretch of that length in the market’s history and compare it with the stretch right after.
Why it matters: a rule can only make money from timing if the market has some memory for it to read. These bars are that memory. Where they sit inside the band, there is nothing to time. How memory is measured
Does last week’s move in one market say anything about next week in another? Each row is the market that moved first.
Of 100 entries, 11 look real on their own; chance alone would produce about 5. None survives being one of 100 tries. The strongest pattern in the matrix fades by half in about 3 days. Taken together, last week explains only 2% to 5% of any market’s next week.
Pick a row, say BTC. Reading across, each square asks: after BTC had an unusual week, what did this other market do the week after? Blue: it followed the same way. Red: it went the opposite way. Outlined squares would look real if you tested only that one pair; filled squares stay real even after we account for having tested all 100 pairs at once.
| Moved this week ↓Next week → | BTC | ETH | LTC | S&P 500 | Emerging | Treasuries | Gold | Oil | Copper | Sol |
|---|---|---|---|---|---|---|---|---|---|---|
| BTC | +0.23 | +0.21 | ||||||||
| ETH | −0.12 | |||||||||
| LTC | −0.23 | |||||||||
| S&P 500 | +0.14 | −0.16 | +0.14 | |||||||
| Emerging | ||||||||||
| Treasuries | −0.12 | |||||||||
| Gold | ||||||||||
| Oil | −0.13 | −0.10 | ||||||||
| Copper | −0.14 | |||||||||
| Sol |
The full equation and the tests
529 weeks of closes from August 27, 2016. Hover a square for its number.
529 weeks of closes from August 27, 2016.
PriceTape & The Committee
PriceTape, today’s incumbent: all cash.
The Committee, the weighted vote of the 23 hand-written rules, is more willing: 36% invested, most of it in ether (15%).
Bitcoin fell 3.2% at the last close.
Both publish their positions every night, win or lose.
PriceTape is this paper’s own trading algorithm: a complex program that writes and rewrites its own rules every night. The Committee is a weighted vote of the 23 hand-written rules that trusts whichever have been making money lately. Both trade crypto and publish their positions every night, win or lose. Below, their positions at the last close, as a share of the portfolio.
PriceTape is named for the ticker tape that once printed every price as it happened.
As of Oct 22
As of Oct 22
Every night a GitHub Action fetches the closes, reruns the board and writes an entry here. A square is a night; an empty one is a night it did not run. 2 of the last 2 nights ran.