Abstract: The study of the traders’ behavioral characteristics was carried out using the methods of experimental economics and machine learning technologies. In order to analyze the impact of level of reasoning on the traders’ profitability, we collected information on cryptocurrencies purchase and sale transactions with a volume of about 715 000 records. Traders, who made more than 15 transactions during the analyzed period, were invited to take part in «Guessing Game». Based on the traders’ answers and information on their transactions, five types of traders were identified, in particular: «Jokers», «Breakeven traders of the 0th level», «Unprofitable traders of the 1st level», «Profitable traders of the 3rd level» and «Professionals».

Analysis of the results showed that, in general, more far‐sighted traders make greater profit. At the same time, «Jokers» are also characterized by a wide variety of strategies, including profitable. Despite the fact that the average guess of «Professionals» was close to the winning value, the variance of guesses for this cluster turned out to be very large. This result indirectly confirms Keynes’s hypothesis that professionals tend to irrational behavior not less than other market agents.

The proposed research actualizes the problems of developing theoretical and empirical models, which are taking into account the degree of traders’ rationality for predicting the cryptocurrency market.