USD/HKD seasonality

Kresmion builds every pair from the ECB euro reference rates of the same day (a pair without the euro is the ratio of the two euro rates), takes the last published rate of each calendar month, and computes each month's return against the previous month-end. For each calendar month it reports the mean, median and share of years with a rise across every completed year since 1999, and t = mean divided by its standard error; an average with |t| under 2 is marked as within noise. The month in progress is shown separately and is never part of an average.

• Every pair uses the ECB euro reference rates (one fixing a day, around 14:10 CET). A pair without the euro, such as USD/JPY, is the ratio of the two euro rates of the same day, so every pair shares one fixing time. History starts with the euro on 4 January 1999; a currency the ECB began publishing later starts later, and the "Since" column shows where.
• A month's return compares the last published rate of the month with the last rate of the previous month. Only consecutive completed months count; the month in progress is shown on its own ("so far") and is never part of an average.
• Per calendar month: the mean and median across years, how many years rose, and t = mean divided by its standard error (the standard deviation across years over the square root of the number of years). An average with |t| under 2 is within noise: it cannot be told apart from zero with this many years. Months with fewer than 3 years show "--".
• With 372 pair-month averages, about one in twenty clears |t| of 2 by chance alone, so a single ringed tile is weak evidence. Calendar averages describe the past; they are not forecasts and not trading advice.