PSX Base Rates: The Yardstick Every Stock Screen Must Beat

ZEME AI Research · measured 2026-08-17 on ~10 years of PSX prices, liquidity-filtered at ≥Rs 5mn/day median 20-day turnover · published 2026-08-22

Any claim about a stock screen — "these names are ready to fly", "this setup doubles your odds" — is meaningless without the base rate it must beat. So we measured the base rates for the liquid Pakistan Stock Exchange universe, next-bar-forward with no lookahead, and we publish them here. If someone quotes you a hit rate, divide it by these.

The yardsticks

horizon	n	median return	P(gain)	P(≥5%)	P(upper lock)
1 session	117,930	—	—	6.83%	3.96%
5 sessions	117,125	−0.17%	48.2%	18.8%	—

Two things worth staring at: the typical liquid PSX stock's 5-session median is slightly negative, and roughly 1 in 25 locks upper circuit on any given day. A screen bragging "40% of our picks locked" is a 10x claim — demand the sample size. Note also that quoting the unfiltered universe (all 485 symbols) flatters P(≥5%) by about a point by including names you cannot actually trade at size.

One session ahead: momentum clustering is the strongest measured signal

setup (all next-bar-forward)	n	P(lock)	lift	P(≥5%)
locked today + 2+ locks in 5 days	2,035	34.7%	8.76x	40.7%
locked today + volume ignited (≥1.8x)	2,095	26.8%	6.77x	33.7%
at 60d high + volatility ≥4%/day	2,257	25.0%	6.32x	31.6%
locked today	4,817	24.3%	6.12x	31.0%
2+ locks in 5 sessions	4,653	22.5%	5.67x	27.9%
at 60d high (<2% away)	6,781	11.6%	2.93x	16.5%
cooled 3–8 days after a lock	17,229	5.0%	1.27x	8.8%

Before anyone trades that top row: a stock that locked today usually cannot be bought (it is at the cap, queued), and buying into an upper lock measured, over 14,831 historical trades, a 48% win rate and a ~0% median — a coin flip. High P(lock) and high tradeability rarely coexist. That tension is the honest headline, not the 8.76x.

The anti-signal: the "coiled spring" measures HALF of random

setup	n	P(lock)	lift
volatility compression (ATR5/ATR20 < 0.6)	7,823	2.11%	0.53x
compressed + near 60d high	699	2.15%	0.54x

Volatility contraction — the classic "coiled spring about to explode" — does not precede a PSX upper lock. It anti-predicts it. A quiet chart is quiet because nothing is happening.

Five sessions ahead: the move is real, and holders give it back

The best 5-session median of any setup we measured is +0.93%. Not one clears 1%. Meanwhile the median maximum favorable excursion of the better setups is +8–9% — the move happens, and holding to day 5 gives nearly all of it back. Typical adverse excursion is −5 to −8%, so inside one week the honest reward-to-risk is roughly 1:1.

setup	n	median (5d close)	P(≥5%) touch	median best point (MFE)
gap-UP open ≥+5%	2,116	+0.93%	34.7%	+8.4%
just locked (day 0)	4,798	+0.78%	34.6%	+9.3%
gap-DOWN open −5..−10%	988	+0.74%	33.2%	+8.1%
already ran >50% in 21d	3,866	−1.14%	31.2%	+8.6%
fell >10% in 5d (dip-buy)	5,312	+0.26%	27.2%	+6.7%

Read the fourth row twice: "already ran >50%" has a healthy-looking 1.66x lift on touching +5% — and a negative median. Ranking screens on touch-probability alone selects high-variance losers. Rank on the median; read the touch odds beside it.

The practical conclusion from the MFE gap: a one-week PSX hold should exit on a price target, not on a date. The +8% typical best point exists; the +0.9% day-5 close is what date-based holding turns it into. And a stop tighter than ~6% sits inside ordinary noise.

Two honest footnotes from building screens on these numbers

A lock-counting screen must check the sign of the current bar. A real case: a stock surfaced in the "2+ locks in 5 days" bucket while its latest bar was a −10% LOWER circuit on 6x volume — prior up-locks matched the rule, and a crash presented as a launchpad.
Entry slip is a selection criterion. A name that closes exactly at its cap is queued and cannot be filled at the reference price; measured on one of our own lanes this cut "50.5% accurate" to 41.9% actually tradeable. Prefer names that closed off their high, even at a lower nominal probability.

How to read this page. These are measured historical frequencies on the Pakistan Stock Exchange, with sample sizes attached, published as the yardstick for evaluating any PSX screen — including our own. They are not predictions or advice, and every number here is era-dependent: regimes change and take the rates with them.

Related ZEME AI research: Do technical indicators work on PSX? · The gap-down open study

ZEME AI · zeme.app · Educational analysis only. Not investment advice. Methodology: next-bar-forward measurement, no lookahead, liquidity filter ≥Rs 5mn/day median 20-day turnover, repaired price history.