Can you tell the difference between real and ‘fake’ stock prices? New study says that most people can

There’s a neat online video game that tests whether you can identify real stock data from randomized, generated data. The game is called ARORA, an abbreviation for “a random or real array” of prices.

(found via Technology Review)

The purpose of the game

Many economists argue that markets are efficient, meaning prices reflect all publicly available information. This implies prices cannot be forecast and will be random and unpredictable like a coin toss. But is this really true?

To find out, the game ARORA was developed and analyzed by researchers Jasmina Hasanhodzic at AlphaSimplex, Andrew W. Lo at MIT Sloan Management, and Emanuele Viola at Northeastern University.

The setup

The goal of the game is to identify the real data between two sets of pictures that display prices.

The real data comes from historical prices of U.S. indices like the Dow Jones Industrial Average or the U.S. spot price of gold.

The ‘fake’ data is a randomized version of the real data. These prices are created by randomly reordering (permuting) the stock returns of historical data. This method keeps the price structure in tact but destroys the time ordering of the data. The beauty of this method is the new series will have the same mean return and standard deviation!

The only thing disrupted is the timing of the returns which is completely randomized, and hence past returns cannot be used to predict future returns.

The gameplay

Basically you are shown two pictures of stock prices. One is real, historical data and the other is a randomized version. You click on a picture with your guess and the game indicates whether your pick was correct or wrong.

You have to make several picks to finish a game.

The study results

The study found that most people could identify the real data better than at random. This was true for all eight tests–the highest p-value was a mere 0.503%! In six of the tests, the p-value was less than 0.001%. Clearly the participants were very good at identifying real data.

The authors suggest the reason is the instant feedback of the game helps the eye spot patterns to distinguish real from randomized data. This skill perhaps relies on our visual recognition system which is something computers cannot mimic yet.

Playing the game

I tested my own skill in the game. I logged in as a guest and then tried the training for the “Lynx” game.

I was a bit disappointed in my results. Out of 10 levels, I was only able to get 6 correct. It’s a little better than half but not by much. But I tried again and I got 7 out of 10 right–I was beginning to see the pattern!

I would imagine someone with a trading or finance background may do well right off the bat. But after some practice things level out–the study did not find any performance difference between those who had finance experience and those who did not.

If you’re up for a challenge, play the game and let me know your experience. How well did you do?

Full study: Is it Real or Randomized? A Financial Turing Test (pdf)