Résultats de recherche pour « Technical trading rules »

We test whether weak-form efficiency in cryptocurrency markets depends on the size of a token. We use daily prices of 55 large-capitalisation and 55 small-capitalisation tokens (without Bitcoin and Ether) from January 2020 to March 2024. For each token we test randomness, autocorrelation, conditional heteroskedasticity, random-walk behaviour and long memory. We compare the groups with exact and rank tests, controlling for multiple testing. We also check whether cross-validated classifiers can find a token's size group from its full set of test results, its efficiency profile. Both groups depart from a random walk, but small tokens depart more. The automatic variance ratio test rejects the random walk for 25 small and 14 large tokens over the full period, and for 24 and 4 tokens in 2022-2024. The efficiency profiles identify the size group out of sample (area under the curve 0.75 for logistic regression and 0.84 for a random forest; permutation p ≤ 0.01). The difference comes from the random-walk and randomness tests, while volatility clustering is common to both groups and long memory is stronger for large tokens. Moving-average and filter rules give larger gross gains on a small-token index, but the gain per day, which is a lower bound on the break-even trading cost, is only 1.3-2.3 times as high. This is probably less than the extra cost of trading the smallest tokens. The results hold without stablecoins and wrapped Bitcoin and agree with the adaptive markets view that inefficiency can survive where trading is expensive.