摘要：This paper points out that even distinct patterns in financial time series, which persist over long periods of time, cannot immediately be taken as genuine. In view of the large number of possible patterns, the only way to avoid any data- snooping bias is to use a formal statistical test, which has not been tailored to the specific patterns present in the data. Adopting a universal frequency domain test for the detection of synchronous cycles, we find clear evidence for within-month patterns in daily returns on the S&P 500 index, which corroborates earlier findings obtained simply by comparing different days of the month.
