Almanac methodology

The almanac answers one question: does the index behave differently at certain points of the calendar than it does on all other days. An effect is a set of windows — “every January”, “the five trading days before each Bank of Russia meeting”. For each window the index return is taken from the close before the window to the close of its last day.

The mean, the median and the share of gains are computed from those returns. The interval next to the mean shows where it lies with 95% probability given the spread: if the interval covers zero, the history cannot confidently name even the direction.

Markets rise without any calendar, so a positive mean alone says nothing. The “Above normal” column is the window return less what the index earns on average over the same number of ordinary days. Ordinary days are all trading days outside the windows of that effect. It is computed in log returns.

The t-statistic shows how far the difference from ordinary days exceeds its own spread; it also accounts for the uncertainty of the ordinary-day estimate itself. p is the probability of seeing such a difference by chance.

The almanac holds several dozen effects, and at a 5% threshold two or three of them would look “significant” by luck. So an effect is called significant only if it survives the Benjamini–Hochberg correction for multiple comparisons; the correction runs across all effects of both indices at once. An effect that looks significant only before the correction is marked separately: it is a reason to look closer, not a pattern.

An effect with fewer than twenty observations is left out of the correction and is never called significant: its t and p are shown for reference, and it is marked “too few observations” — that means “we do not know”, not “not significant”.

• Month of the year — the whole calendar month; an unfinished month forms no window.
• Day of the week — the return of one trading day against the previous close. Saturday and Sunday are rare working sessions of past years.
• Bank of Russia meeting — the five trading days before the decision day, the decision day itself and the five trading days after it; separately, the days after the decision depending on whether the rate was cut, held or raised.
• Quarter turn — the last and the first five trading days of a quarter.
• Tax period — the five trading days ending on the tax payment day: the 28th or the first trading day after it.

The MOEX Russia Index is a price index: dividends are not included. In the payout season it loses the size of the dividend gaps, so summer months look weaker than they were for an investor. The RGBITR government bond index is a total return index and includes coupons.

Only closed trading days enter the calculation. A window counts once it is finished: five trading days must pass after a meeting, a month must end.

A record of the past does not promise a repeat. This is not a forecast and not investment advice.

The base is the close of the last day the stock traded with the right to the dividend. The gap is closed on the first day the close returns to the base; the duration is the number of trading days from the base day. Payouts yielding less than 1% are left out: they do not form a gap.

A fresh gap that has not closed yet can be neither dropped nor counted as closed — either would overstate the speed. So the median and the share closed within 30 days come from the Kaplan–Meier estimate, where such a gap is an unfinished observation. A gap is observed for 250 trading days; prices are adjusted to today's share count so that a split does not read as a gap. For a stock with fewer than five payouts no statistics are computed.