Skinner-box rats trained to predict currency market movements


Viennese artist Michael Marcovici's Rat Traders uses reward, punishment and selective breeding to create a strain of lab-rat that can predict the movement of international currency markets.

The rats are kept in Skinner boxes, and are trained by playing them piano scores whose pitch reflects actual historical currency market fluctuations; when the music ends, each rat chooses a door, one representing a long position, the other a short one. If they correctly anticipate the market, they're given food; for incorrect predictions they get a shock. The most successful rats are bred to produce new generations.

The outcome is a group of rats that can predict USD/EUR market movement with 53% accuracy.

Or, possibly, the result is a bunch of credulous posts like this one, accepting the reality of the experiment without much evidence, but producing an attention-bubble that says that high-flying finance operatives underperform relative to lightly trained rats, which is also a useful contribution to the literature, albeit not the scientific literature.

In any event, Skinnerian training doesn't usually involve punishment for undesirable behaviors; the point of "positive reinforcement" is to reward "good" behaviors and ignore everything else.

The next step was the rats' training. I produced about 800 different ticker tracks of different market situations. Since I did not want to render the whole story too complicated, I only used the USD/EUR future to turn the rats into experts in this specific market segment: other rats though may be educated in other markets as well. The training took about three months. I started with 80 Sprague Dawley laboratory rats, 40 males and 40 females, with the intention to cross the best of them to genetically create the best traders through select breeding. The training environment was a so called Skinner Box, widely used in experiments and industry for behavioral experiments with animals. The rats were separately trained for five hours daily (thanks to Anna, Gerda, and Dirk who did a great job in the past months). Every day the rats were confronted with 100 different ticker tracks; the goal was making them seek out sound-patterns that humans are not able to recognize and predicting the next market move after the last sound heard. ( I am currently working on a website enlisting human training programs as well). Each time after listening to a sound, the rat had to choose between pressing either a green or a red button, green for "long" (if the prices were expected to move up), red for "short" (if they predicted a decline in prices). When they were right they received a small amount of food (the good rats became fat very fast); when they took the wrong button, they received a minor electric shock. Very soon it showed that some rats were doing outstandingly well: they developed a good ability to remember the patterns they were listening to; we needed them to react to real time data though. To get there, we developed the following training cycle:

week 1 training with historic ticker tracks 1 to 200
week 2 training with historic ticker tracks 201 to 400
week 3 training with historic ticker tracks 401 to 600
week 4 training with historic ticker tracks 1 to 200
week 5 training with historic ticker tracks 201 to 400
week 6 training with historic ticker tracks 401 to 600
at this point the rats from the first round that did not improve were sorted out; 32 rats stayed.
week 7 training with historic ticker tracks 1 to 200
week 8 training with historic ticker tracks 201 to 400
week 9 training with historic ticker tracks 401 to 600
again we excluded the non-improving candidates from the training group; now we had 22 rats left.
week 10 training with historic ticker tracks 601 to 700
this was the real test now: the future traders had not heard the tracks before. We had to sort out another 6 rats
week 12 training and testing with historic ticker tracks 701 to 800 without repeating any of the tracks during the training
we were now able to make out the really gifted traders and eventually ended the test series with 4 reliable rats:

Mss. Kleinworth
Mss. Coutts
Mr. Morgan
Mr. Lehmann (we were quite surprised to find he made it)

…the training was almost finished; the performances of the top 4 rats had turned out to be comparable to those of the world' s best fund managers. Their ability to recognize sound patterns generated from the market's ticker tape was incredible. And the rat traders also hold another advantage: unlike their human counterparts, they are not likely to be distracted from news or economy fundamentals, their own personal or their bank's financial status.

Rat Traders

(via Making Light)