Do Technical Indicators Work on the Pakistan Stock Exchange? We Tested 12 of Them on 512,118 Observations

ZEME AI Research · study completed 2026-08-20 · liquid PSX universe, ~10 years of repaired price history · published 2026-08-22

MACD crosses, the golden cross, price crossing EMA50, Bollinger breakouts, stochastics, EMA-stack alignment, Ichimoku, ADX ignition and OBV divergence — the classical technical-analysis toolkit — tested as events (the day each condition becomes true), entered at the signal close, measured 5 and 20 sessions forward, on 512,118 liquid PSX symbol-days. Each signal was era-split by its own median date: a verdict HOLDS only if the 5-session median beats that era's own base rate in both halves. That out-of-sample discipline is what separates a real edge from a fitted one.

The headline: 12 classical signals, zero economically alive

signal	n	early era (med5 vs base)	late era (med5 vs base)	win 5d	verdict
MACD cross	21,184	−0.32 vs −0.32	−0.26 vs −0.12	46.5%	FAILS
golden cross (EMA50×200)	1,703	−0.18 vs −0.33	−0.44 vs −0.09	47.2%	FAILS
price crosses above EMA50	23,683	−0.68 vs −0.33	−0.38 vs −0.11	44.5%	FAILS
Bollinger break up	19,910	−0.09 vs −0.33	+0.00 vs −0.09	49.1%	"holds", median 0.00%
Bollinger squeeze-break	6,861	+0.00 vs −0.30	−0.21 vs −0.13	48.7%	FAILS
stochastic oversold cross	14,754	−0.28 vs −0.30	−0.35 vs −0.15	46.0%	FAILS
stochastic >80 entry	16,630	−0.22 vs −0.30	−0.12 vs −0.14	48.0%	"holds", median negative
EMA20>EMA50 stack	21,205	−0.57 vs −0.33	−0.50 vs −0.09	44.5%	FAILS
Ichimoku tenkan/kijun cross	13,549	−0.17 vs −0.34	−0.13 vs −0.13	47.7%	"holds", median negative
Ichimoku cloud break	14,406	−0.70 vs −0.37	−0.36 vs −0.11	—	FAILS
ADX>25 +DI ignition	9,968	−0.55 vs −0.37	−0.31 vs −0.11	—	FAILS
OBV divergence (quiet accumulation)	12,926	−0.49 vs −0.26	−0.33 vs −0.22	—	FAILS, med20 −1.23%

The two signals that technically "hold" against their base rate — Bollinger break-up and stochastic-overbought entry — beat it by roughly 0.1–0.2 percentage points with an absolute median of 0.00% and a sub-50% win rate. After spread and fees, that is a loss. Statistically alive, economically dead.

The most trusted signals are the worst ones

Buying an EMA20>EMA50 "aligned uptrend" measured −0.50% vs a −0.09% base — half a point worse than a random pick. Price crossing above EMA50: −0.38 vs −0.11. The golden cross fails in the recent era at −0.44 vs −0.09. Classical trend-following entries on PSX are systematically the wrong side of the trade. By the time the chart "confirms," the confirmation buyer is the exit liquidity for whoever accumulated earlier.

The OBV finding deserves its own line. OBV rising while price stays flat is the textbook proxy for quiet accumulation. On PSX it is not just edgeless but harmful: −1.23% at 20 sessions, the worst 20-day result of all 12 signals. Volume-flow divergence here picks up distribution disguised as accumulation.

Market regime makes everything worse, not better

Conditioned on the KSE-100 being above or below its own EMA50 on signal day: every one of the 12 signals degrades in a down-regime, most by a lot (Bollinger break: −0.12 in an up-regime → −1.22 down; squeeze-break −1.58 down). No classical signal is a down-market shelter.

What survived: two behavioral effects, not indicators

The same harness run over 10 further behavioral setups (22 signals total) produced exactly two with positive medians in both eras:

Fresh 52-week-high close (n=8,897): median +0.12/+0.28 across eras, 20-session median +0.37%, and the equal-best measured 5-day odds of touching +10% (34.1%). The structural reason: a fresh high has no overhead supply — nobody is trapped above, waiting to sell into the move. Strictly regime-dependent: +0.40 median in an up-regime, −1.11 in a down-regime.
Month-end effect (n=23,186): buying the last session of the month and holding into the new month measured +0.77% vs a −0.11% base in the recent era, 52.5% win — consistent with institutional new-month flows.

How to read this page. These are measured historical frequencies on the Pakistan Stock Exchange, published so the numbers are on the record. They are not trade recommendations, and a pattern that held for ten years can stop holding. Anything with a 0.00% median loses money after costs. If a strategy is sold to you on MACD, moving-average crosses or Bollinger breakouts on PSX — ask for the era-split table.

Related ZEME AI research: The gap-down open study · PSX base rates — the yardstick

ZEME AI · zeme.app · Educational analysis only. Not investment advice. Measured on repaired PSX price history with quarantined symbols excluded; methodology: event-based entries, era-split out-of-sample validation, per-era base-rate comparison.