Glossary

Scan, Review and Check

The figures of the Scan, a rule's Review and Check, all from one evaluator. Every figure is gross and compares each trade with holding the same symbol for the same length. No figure is a verdict: the Scan's filters are where you decide what is enough.

Average return

The mean gross return of every trade the rule counted, from entry to exit, in its direction.

Same-holding return

What holding the same symbol, in the same direction, for the same number of sessions made on average, over every such hold before the research cutoff. It is what the trade would have earned without any calendar idea.For a short, it is the rise the short has to overcome, so it is usually negative for the trade. The registered strategies call it an ordinary same-length hold, and older research documents call it drift.

Average excess

Each trade’s return minus its same-holding return, averaged over every counted trade. Above zero, the calendar added something beyond the symbol simply moving.The registered strategies’ studies and the research documents call it the edge over holding, or the excess over drift.

Median excess

The middle trade’s excess: half the trades beat holding by more, half by less. Far below the average excess means a few large trades carry the average.

Excess / year

Each year’s total excess, the sum of that year’s trade excesses, averaged over the years observed. It puts a rule that trades once a year and one that trades every week on the same one-year footing.The Scan lists rules by it, largest first, unless you sort by another column.

Trades / year

The counted trades divided by the years observed.

Years

The calendar years with at least one counted trade. For a weekday, week-of-month or month-end rule, only years the symbol’s prices cover in full count, so a partial first or last year never skews a total.

Positive years

The share of observed years whose average trade beat holding, a year mean excess above zero. Around half is what chance gives.

Rolling 4Y positive

The share of four-consecutive-year windows whose average yearly excess was above zero. A missing year breaks a window.The windows overlap, so they are not independent: it measures how steady the rule was, not how likely it is to work.

Latest 4Y

The average yearly excess over the four years ending in the research cutoff’s year. Empty when any of the four is missing.

Last 2Y

The total excess of the research cutoff’s year plus the year before it. Empty unless both years are observed.

Without best year

The average yearly excess with the single best year left out. If it falls to near zero, one year made the rule.

Worst year

The year whose trades did worst against holding, by their average excess, and that figure.

Next occurrence

The rule’s next entry and exit on the exchange’s calendar, from today on, and how many sessions remain before the entry.

Since the cutoff

What the rule did after the research cutoff: the sum of the gross returns of every trade that has closed since, with how many there were. Out of sample: none of these trades reached a figure the rule was found or ranked by.The research figures stop at the cutoff, the end of the last completed year, so this is the current year. The excess beside it compares each trade with the same-length hold as it was known at the cutoff.

Rule families

The seven kinds of calendar rule the Scan and Check evaluate.

Day of week

Enter on one weekday and exit on another; an exit weekday at or before the entry weekday means the following week. Many trades a year.

Month-end

Enter and exit a fixed number of sessions before or after the last session of each month, from five before it to five after.

Week of month

Enter on the Nth Monday of the month and exit on the next Monday, a week later, which may fall in the next month. A month with no Nth Monday has no trade.

Day of year

Enter at the first session on or after a calendar date and hold for a fixed number of sessions: 5, 10, 21 or 42 in the Scan.

Week of year

One of 52 week buckets counted from 1 January: week 1 starts on 1 January, week 2 on 8 January. Enter at the first session of the bucket and exit at the first session of the next.Week 52 runs to the first session of the new year, so 31 December belongs to it.

Month of year

One calendar month: enter at the first session of the month and exit at the first session of the next.

Event

Sessions counted from a dated event: a scheduled FOMC announcement, or a weekday the exchange is closed. Enter at the open of one session near the event and exit at the open of a later one.The event session is the announcement day itself, or the first session after a holiday, so the pre-holiday session is one session before it. The dates are known in advance: the Federal Reserve publishes its calendar a year or more ahead.

Returns and risk

Gross per trade

If a round trip costs you more than this, the average trade loses money.What a trade made from its entry price to its exit price. Every return in the product is gross: no transaction cost is charged to it (research contract rule 7). Cost belongs to the instrument, the order and the broker, not to the calendar; enter your own round trip on the Strategies page to see which strategies would not pay for it.

Break-even cost

The round-trip cost at which the average trade stops paying: its gross return per trade.SPY or ES bought and sold at the close typically costs 1 to 3 bps a round trip; a single stock at the open costs more.

Drift

The tendency of a stock to rise over time. Most seasonal windows look good only because the stock went up in every window, and drift is that rise. The same-holding return measures it.

Win rate

The share of trades that made money. On a rising stock almost any long window wins most of the time, so a high win rate alone says little.Called hit rate on the registered strategies’ pages.

CAGR

Compound annual growth rate: the steady yearly return that would turn the starting money into the ending money over the whole span.

Return per year in the market

CAGR scaled to the time the money was actually at risk. A window in the market 3 percent of the time cannot match holding on raw CAGR; this figure compares like with like.Computed as the strategy’s total growth compounded over the years it spent in the market rather than over the whole span. The rest of the time the money is free for something else.

Max drawdown

The worst fall from a previous high, marked daily, before a new high was reached. It is the pain you would have had to sit through.

Time in the market

The share of trading sessions in which the strategy held a position.

Profit factor

Money made on winning trades divided by money lost on losing trades. Above 1 the strategy made money overall.

Evidence

p-value

How often luck alone would produce an edge this large if the window had no real effect. p = 0.16 means about 1 time in 6: not evidence. p below 0.01 means under 1 in 100.It is tested on the edge over holding, not on the raw return, so the stock simply going up does not count as evidence.

q-value

A p-value corrected for how many windows were searched. If you look at thousands of windows some will look great by luck; q is the share of results this strong you should expect to be luck.Also called the false discovery rate (Benjamini-Hochberg).

Allowing for how many were tried

Finding the best of many windows by looking is much easier than predicting one in advance. The bar a result must clear rises with the number of windows searched.

Out of sample

Years the rule never saw while it was being chosen. A result that holds there is harder to fake than one measured on the years that picked it.

Walk-forward

Choose the rule using only past years, trade it the next year, then move one year on and repeat. It imitates what a trader could actually have done.

Research cutoff

The last date the research was allowed to see. The Scan, a Review and Check use prices through it; everything after it is untouched, so trades after the cutoff are a genuine test.

Catalog labels

Given to the old catalog's windows, which Track Record still evaluates on held-out prices. The Scan gives a rule no label.

Lead

A catalog row surfaced by scanning for you to investigate, never a signal to trade.

High confidence

Beats holding even after allowing for the search, kept at least 40 percent of its edge on years it was not chosen on, and has not weakened recently.

Validated

Beats holding even after allowing for how many windows were searched (q below the bar).

Promising

Worth watching, not proven. At least 10 years, p below 0.01 and a win rate of 55 percent or more, but it does not survive the allowance for how many windows were searched.

Exploratory

Too few years (7 to 9) or weak statistics. A lead to look at, not to trade.

Reject

Too few years, a non-positive average, or broken in recent years. Not shown in the lists.

Strategy labels