Sequential filtering and Top-N ranking

Also seen as: filter order, the filter funnel, Top-N, best N

How does a screener apply its filters?

A screener starts from its universe and hands that set to its first filter. The instruments that pass become the input to the second filter, those survivors become the input to the third, and so on to the end of the list. What comes out of the last filter is the set of matches. That set is also what a screener backtest measures.

Because every filter has to be satisfied, the filters combine with AND and the count can only fall or stay level as you go down the list — a funnel, never a fork. If one filter leaves nothing, the screener has no matches and the filters after it have nothing left to narrow.

Why does filter order matter?

Order matters because Highest and Lowest rank the survivors at their own position, not the universe. Move a ranking filter and you change the pool it ranks, so you change which instruments come out — even though the rules themselves are identical.

Threshold filters — <, ≤, =, ≥, >, Between — are order-independent among themselves: an instrument either clears a threshold or it does not, whatever ran before. Put one of them before a ranking filter and it shrinks the pool the rank draws from; put it after and it trims the ranked set. Both are legitimate screens; they are different screens.

What does a Highest or Lowest filter keep?

A Highest or Lowest filter keeps at most N instruments — the count you set, taken from the instruments still passing at that point. It can return fewer than N in two situations: when fewer than N instruments are still in the running, and when some of the survivors have no reading for that metric, because an instrument that cannot be measured cannot be ranked and drops out.

Note what Top-N does not say: it selects by relative position, not by level. The "highest 10" on a metric are the highest of that particular pool, which says nothing about whether their readings are high in absolute terms.

Defaults in Fincanva

• Filters run in the order they sit in the filter strip, which is the order you added them; the strip has no reorder control today, so changing the order means removing a filter and adding it again.
• A ranking filter's count defaults to 5, and its stop list runs 1 to 15 one at a time, then in widening steps up to 300.
• Highest and Lowest sit in the Rank group of a filter's operator dropdown, and only filters whose catalogue entry declares them offer them. See standard operators.
• A ranking filter reads the same metric as any other filter on that metric, including its Advanced settings, so "Highest 10 by revenue four reports ago" is a valid rank.
• The match count reflects the end of the whole sequence, not any intermediate stage. A strategy fed by that screener may work with fewer candidates than the count shows — see max-symbols cap.

Worked example: the same two rules in both orders

A screener over 500 US stocks with two filters: P/E < 20 and Highest 10 on ROE.

Order A — threshold first, then rank.

The result is the 10 most profitable stocks among the cheaper half — every one of them trades below 20× earnings.

Order B — rank first, then threshold.

The rank now runs across all 500 stocks, so it returns the market's 10 most profitable companies — which tend to be expensive. The P/E filter then cuts that list of 10 down to the 3 that happen to trade below 20× earnings.

Same two rules, same universe, same day: 10 instruments one way and 3 the other, with a completely different character. Order A asks "who is most profitable among the cheap ones"; Order B asks "of the most profitable companies, which are cheap". Reading a ranking filter's position as part of the rule, not as cosmetic list order, is the habit that stops this surprising you.