Sab se mehnga aqeeda jo humne test kiya
Aqeeda ye hai: wo stocks dhoondo jin ka operator pehle jeetta raha ho - yani campaign success rate ooncha ho - aur us operator ke peeche chalo, kyunke acha operator acha hi rehta hai.
Is ko seedha measure kiya ja sakta hai. Panel ke har us stock ke saath jis ka operator history hai, us operator ka historical success rate bhi mojood hai. Un ko paanch barabar groups mein baant kar dekhte hain ke agle 20 sessions mein stock ne kya kiya:
| operator ka purana win rate | agle 20 sessions ka average return | n |
|---|---|---|
| top paanchwan hissa (behtareen record) | +1.272% | 120,903 |
| neeche wala paanchwan hissa (bad-tareen record) | +2.264% | 126,765 |
Behtareen minus bad-tareen = -0.99 pp, aur t = -13.84, 247,668 stock-sessions par.
Pehle nishaan (sign) parhein, phir size
t agar taqreeban +/-2 se bahar ho to ittefaq hone ka imkaan kam hai. Ye us se bohot door hai - aur negative hai. Yani baat ye nahi ke "operator ka record kaam nahi karta" - us soorat mein t sifar ke aas paas hota. Baat ye hai ke jin stocks ke operator ka record behtareen tha, unhon ne aage ja kar kam kamaya, aur ye farq itna bara aur itna consistent hai ke qismat nahi ho sakta.
Aisa kyun hota hai?
Sach ye hai ke ye lesson kya hua wo measure karta hai, kyun hua wo nahi. Lekin sab se mumkin wajah koi raaz nahi: shandaar record khud is baat ka saboot hai ke move pehle hi ho chuka hai. Operator ka success rate kamyab campaigns ke baad barhta hai. Jab tak record acha dikhne lagta hai, aap late ho chuke hote hain. Record peeche dekhne wala sheesha hai jise log windscreen samajh kar parhte hain.
Ye lesson ye NAHI kehta
Ye ye nahi kehta ke ab bad-tareen record wale operators khareedo. Neeche wale group ka return positive zaroor hai lekin wo bhi round-trip cost cross nahi karta (lesson 3 dekhein), aur ek toote hue signal ko ulta kar ke istemal karna noise fit karne ka purana tareeqa hai.
Sahi nateeja is se tang aur zyada kaam ka hai: operator ka historical win rate kisi aage ke faisle ka input nahi hona chahiye - aur jahan ZEME ki apni surfaces us ko chhod deti hain, wo isi wajah se chhodti hain, aur ye baat likh kar batati hain.
Evidence - History - Confidence - Counterargument
Confidence - ye ke asar asli aur negative hai: ooncha. Ye ke koi is ke ulta par amal kare: bohot kam.
Counterargument - sab se mazboot aitraaz: top quintile mein shayad wohi stocks bhare hue hain jo pehle hi bhaag chuke hain. To ho sakta hai ye sirf run ke baad ki mean-reversion measure kar raha ho, operators ke baare mein kuch bhi nahi. Ye ek asli mutabadil wajah hai - aur ittefaq se nateeja wohi rehta hai: record ko aage ka input mat banao.
Ye ZEME Academy ka taleemi tajziya hai - investment advice nahi, koi recommendation nahi. Har number guzri hui history hai, aur history peshangoi nahi hoti.
Saboot - har number ka source
Is lesson ka har number yahan apni definition, apne sample size, apni window aur us exact query ke saath mojood hai jis se wo bana. Aap khud dobara chala kar dekh sakte hain.
opwr_top_ret = +1.272% (n = 120,903)
- what it is: mean 20-session forward return of the stocks whose operator has the BEST historical win rate (top quintile)
- window: 2016-01-01..2026-07-30
- source: zeme_uat_panel.parquet
- how it was computed (pandas over the panel, not SQL):
``text Split liquid rows that carry an operator history (camp_success_rate >0, a PERCENTAGE 0-100) into quintiles of camp_success_rate; compare mean fwd_ret_20_e0 of the top quintile against the bottom; Welch t. ``
opwr_bot_ret = +2.264% (n = 126,765)
- what it is: mean 20-session forward return of the stocks whose operator has the WORST historical win rate (bottom quintile)
- window: 2016-01-01..2026-07-30
- source: zeme_uat_panel.parquet
- how it was computed (pandas over the panel, not SQL):
``text Split liquid rows that carry an operator history (camp_success_rate >0, a PERCENTAGE 0-100) into quintiles of camp_success_rate; compare mean fwd_ret_20_e0 of the top quintile against the bottom; Welch t. ``
opwr_spread = -0.99 pp (n = 247,668)
- what it is: best-operator minus worst-operator forward return. A POSITIVE number would mean the operator's record predicts the next month; a NEGATIVE one means it predicts it BACKWARDS
- window: 2016-01-01..2026-07-30
- source: zeme_uat_panel.parquet
- how it was computed (pandas over the panel, not SQL):
``text Split liquid rows that carry an operator history (camp_success_rate >0, a PERCENTAGE 0-100) into quintiles of camp_success_rate; compare mean fwd_ret_20_e0 of the top quintile against the bottom; Welch t. ``
opwr_t = -13.84 (n = 247,668)
- what it is: Welch t of that difference. |t| above ~2 is unlikely to be chance
- window: 2016-01-01..2026-07-30
- source: zeme_uat_panel.parquet
- note: sign matters more than size here: a significant NEGATIVE t means the belief is inverted, not merely useless
- how it was computed (pandas over the panel, not SQL):
``text Split liquid rows that carry an operator history (camp_success_rate >0, a PERCENTAGE 0-100) into quintiles of camp_success_rate; compare mean fwd_ret_20_e0 of the top quintile against the bottom; Welch t. ``
Ye ZEME Academy ka taleemi tajziya hai - investment advice nahi, koi recommendation nahi. Har number guzri hui history hai, aur history peshangoi nahi hoti.