Deadbeats, the word is on the street that your Fibonacci chart is out

Receive free Cryptocurrencies updates

We’ll send you a myFT Daily Digest email rounding up the latest Cryptocurrencies news every morning.

Investment banks keep technical analysts on staff for the same reason that health services employ chiropractors. Evidence of efficacy is weak but some customers seem to like it and may strongly believe they’ll benefit, and no good comes from seeking to prove them wrong.

But chartists and chiropractors have the same problem when seeking acceptance by the mainstream: scientific method doesn’t work well on their vibes-based disciplines. If a buy signal or a treatment for lumbago has emerged from one person’s subjective interpretation of selected information within a system of complex interconnections, it’s a tricky thing to replicate in test conditions.

No matter. Doctors and investment bankers are both in the business of making confident predictions about uncertain outcomes. If they think it’ll please the client, where’s the harm in sprinkling in a bit of hocus-pocus?*

Here’s a fun thesis by Kristian Ratia, a computing lecturer at Oulu University of Applied Sciences in Finland and a crypto enthusiast. He fed some crypto token prices into popular technical analysis algorithms then traded only on their strongest buy and sell signals.

Crypto’s a good match for this kind of study. A core tenet of chartism is that all known fundamentals will be in the price, so historic trends reflect marketplace psychology in ways that are likely to echo in the future. But with conventional assets, all sorts of fundamentals will inform supply and demand, many of which can’t be known simultaneously by all parties — a parking lot may be less full than last week, for example, or peace talks might have stalled, or a VIP may have just fallen down a well — and it’s only by deploying a daedal definition of the efficient market hypothesis that we’re permitted to ignore all these possibilities.

Enter shitcoins, the naked mole-rats of financial markets research. Crypto tokens generally lack even the most basic metrics needed for fundamental analysis, such as utility and cash flow. The only fundamental that matters is the likelihood of someone pulling a scam. After that, promoters insist, it’s all trend.

Here are Ratia’s results:

The above needs a lot of context, starting with methodology. The experiment took 22 minor-league crypto tokens and ran Binance-sourced trade data through five trend-detector algorithms.

Bots were set running using various commonly used parameters and each individual build was put into learning mode for 200 days, after which it bought or sold whenever its algorithm generated an “excellent” signal. Trades were in increments of $1,000 from a starting bank of $100,000 and, to keep a lid on the 0.075 per cent-per-transaction dealing cost, trades were limited to a maximum of one per minute.

The experiment ran between the first day of 2022 and April 15, 2023, so was born in interesting times. The crypto universe lost about 70 per cent of its value over the period. Ratia’s trade bots came out of learning mode shortly after Celsius and Three Arrows Capital filed for bankruptcy, then had to navigate FTX’s collapse a few months later.

In theory, the crypto winter shouldn’t matter. “One of the great strengths of technical analysis is its adaptability to virtually any trading medium and time dimension,” writes John J Murphy in the 1986 book Technical Analysis of the Futures Markets.

But full automation can’t do vibes-informed adaptability like John J Murphy. The bots stuck rigidly for the whole period to their assigned classic TA method: Bollinger Bands, Exponential Moving Average (EMA), Moving Average Convergence Divergence (MACD), Relative Strength Index (RSI) or Simple Moving Average (SMA).

What followed were occasional successes and lots of failure:

Would buying and selling at random have been more profitable? Assuming the trader trades with the hyperactivity of an over-optimistic bot, no.

The experiment’s control group involved an algorithm choosing randomly 10,000 times over the test period between buying, selling and doing nothing. This delivered an average performance of negative 52 per cent, which is eight percentage points worse than the chartist average:

We’ve managed this far without naming the tokens involved. The below table has the full shitshow, along with their performance over the study period using the random buy-sell algorithm. Though none has become a household name, and quite a few turned out to be doughnuts, all these coins were sufficiently mainstream to be quoted in a trade pair with BUSD, Binance’s native token:

The research also has a performance review broken down by a token’s purported main function or attribute, which shouldn’t really matter in TA but perhaps it does? Going by the data it’s quite hard to say anything definitive.

(Click on dots for details of which token is in which category, and on the key to toggle the measures on and off.)

What to conclude? Probably nothing. Technical analysis is, after all, all about belief not scientific method. Vibes won’t be constrained by circumstantial evidence, whereas any experiment is doomed to be boxed in by its own methodology.

For his part, Ratia keeps the faith. His interests in crypto and technical analysis stem from the same love of numbers, not money, and all maths problems are there to be solved.

“I am an engineer, so I do not believe in luck, I believe in my server calculation power,” he told Alphaville. “Sometimes you might need less calculation to find a good [trading] strategy, sometimes you need more calculation. If it takes a long time to find a profitable strategy, I will say that you will need a more powerful computer.”

Which is fair enough.

Quantitative finance is to technical analysis as physiotherapy is to chiropathy: one is recognised science and the other is not, but overlaps are such that the line between sense and nonsense is to a large degree just personal preference. Some people will look at the above tables and see only bunk; others will think they sense the shadow of a framework of a system that, with the deployment of more brainpower, could be made profitable. Neither view is demonstrably wrong.

That’s the problem with technical analysis: methodological studies on the ways prices might have memory or mystical properties are only really only good for reinforcing everyone’s priors. We’re sure you’ve heard this all before, but you’ve never really had a doubt.

* Complaints to the usual address. Please include “I’m a chartist”, “I’m a hodler” or “I’m a chiropractor” in the subject line.