LissafiKafin bugawaNazariyyaKaratun minti 5

TAFIYE-TAFIYEN BAZATA BIYU DA SUKE KAMAR SUNA DA ALAƘA

A 1926, wani abu mai tayar da hankali ya ja hankalin Yule: jerin bayanan lokaci (time series) masu cikakken zaman kansu na iya nuna manyan daidaito (correlations). A wurin Yule, sunansa “daidaito marar ma’ana” (nonsense correlation). Waɗannan gwaje-gwajen da tafiye-tafiyen bazata (random walks) — jerin da aka gina ta hanyar tara matakan bazata ɗaya bayan ɗaya — sun nuna daidaito masu warwatsuwa fiye da yadda samfura masu zaman kansu (independent samples) za su nuna nesa ba kusa ba. Ayyukan da suka biyo baya sun tabbatar da cewa waɗannan daidaito na ƙarya (spurious correlations) ba sa ɓacewa idan jerin suka ƙara tsawo.

Liping Zhu da Zhongming Xie, a Cibiyar Ƙididdiga da Babban Bayanai (Institute of Statistics and Big Data) ta Jami’ar Renmin ta China (Renmin University of China), sun tambayi ko matsalar tana tsira daga wani kayan aiki na zamani fiye da haka.

Ma’aunin da bai kamata a yaudare shi ba

Daidaito na yau da kullum (na Pearson) yana gano alaƙar layi madaidaici kawai. Daidaiton nisa (distance correlation), wanda Székely da abokan aikinsu suka bayyana a 2007, yana da babban buri: ga masu sauyawa na bazata biyu, yana zama sifili idan kuma kawai idan (if and only if) suna da zaman kansu, komai siffar alaƙarsu. Koyaushe yana tsakanin 0 da 1.

Amma wannan tabbaci yana ɗauka cewa bayanan zaɓe ne masu zaman kansu (independent draws). Tafiye-tafiyen bazata guda biyu sun sha bamban. Kowane mataki yana shafar kowace ƙima ta gaba ta tafiyar, don haka lura masu bin juna suna raba abin da ya gabata, kuma yaɗuwarsu tana ci gaba da sauyawa da lokaci. Kwaikwayo sun riga sun nuna manyan daidaiton nisa tsakanin tafiye-tafiyen bazata masu zaman kansu, waɗanda marubutan baya suka danganta da dogaro na jere (serial dependence) na tafiye-tafiyen da yaɗuwarsu mai sauyawa; abin da babu shi kuwa shi ne ka’ida game da iyakarsu (limit).

Iyaka da ke tsakanin 0 da 1 tsantsa

Marubutan sun tabbatar da cewa, yayin da tafiye-tafiyen ke ƙara tsawo, daidaiton nisa ba ya raguwa zuwa sifili. Ga kowace tafiya mai matakai masu zaman kansu, masu rarrabuwa iri ɗaya (independent, identically distributed), masu iyakantaccen bambanci (finite variance), yana nufa (converges) zuwa wata ƙima ta bazata da hanyoyin Brown (Brownian paths) biyu ke bayarwa, wato iyakar tafiye-tafiyen bazata mai ci gaba (continuous limit). Wannan iyaka ta duniya ce (universal) — ba ta dogara da yadda matakan suka rarrabu ba — kuma tana tsakanin 0 da 1 tsantsa, tana ɗaukar ƙimomi a duk faɗin wannan kewayo. Don haka damar ganin daidaito sama da kowane mataki (threshold) da aka zaɓa, misali 0.3, ba ta taɓa ɓacewa.

Mabuɗin shi ne a duba tsawon lokacin da hanyoyin biyu ke ɗauka a kowane yanki, idan aka ɗauki samfurinsu a lokaci ɗaya na bai ɗaya. Hanyoyi biyu na bazata (processes) za su iya kasancewa masu zaman kansu a tsakanin aukuwa daban-daban (realisations), yayin da wani biyu-biyu guda da ya auku, idan aka bi shi cikin lokaci, har yanzu yana nuna wani tsarin alaƙa.

Kwaikwayo: a maƙale a 0.5

Ƙungiyar ta kwaikwayi biyu-biyu 2,000 na tafiye-tafiye masu zaman kansu masu matakan Gauss (Gaussian steps) ga kowane tsawo daga matakai 50 zuwa 5,000. Ga matakan da kansu, daidaiton nisa yana dusashewa kamar yadda ake tsammani: daga 0.244 a 50 zuwa 0.025 a 5,000. Ga tafiye-tafiyen da aka gina daga waɗannan matakai guda, yana tsayawa a kusan 0.5 a kowane tsawo.

A matakai 5,000, tsakiyar 90% na biyu-biyun tafiye-tafiye sun bazu daga 0.215 zuwa 0.844, kuma biyu-biyu 1,644 daga cikin 2,000 — 82% — sun wuce 0.3. Marubutan sun jaddada cewa ƙimar da ke kusa da 0.5 sakamako ne na gwaji (empirical finding), ba tabbataccen adadi mai tsayayye (proven constant) ba.

Sassa biyu: a hagu, matsakaicin daidaiton nisa yana tsayawa kusa da 0.5 ga tafiye-tafiyen bazata yayin da yake sauka zuwa sifili ga matakansu; a dama, lanƙwasan rarrabuwa (distribution curves) na tsawon tafiya uku sun kusan rufe juna.

(a) Matsakaicin daidaiton nisa ga tafiye-tafiyen bazata (cikakken layi) da kuma ga matakansu masu zaman kansu (layi mai karya-karya), tare da ɗamarar 5–95%, a kan adadin lura. (b) Rarrabuwar daidaiton tafiye-tafiye ga matakai 100, 1,000 da 5,000; ɗigon da aka yi wa alama yana nuna cewa 17.8% daga cikinsu ba su wuce 0.3 ba a matakai 5,000. — Hoto na 1, Zhu & Xie (2026), arXiv:2611.16614.

Gwajin da ke ƙararrawar ƙarya koyaushe

Hanyar da aka saba bi don gwada zaman kai da daidaiton nisa ita ce jujjuya tsari (shuffle): a cakuɗa jerin lokacin ɗaya daga cikin jerin bayanan sau da yawa, sannan a duba ko ainihin ƙimar ta fice. Marubutan sun tabbatar da cewa ga tafiye-tafiyen bazata, wannan gwaji yana ƙin zaman kai (rejects independence) da yiwuwar da ke kusantar 1, komai matakin kuskure (error level) da aka zaɓa.

Kwaikwayon sun yarda da haka. A matakin 5% da aka saba, gwajin jujjuya tsari ya bayyana alaƙa bisa kuskure ga 94% na biyu-biyun tafiye-tafiye masu zaman kansu na matakai 100, da kuma ga dukkan biyu-biyu 300 na matakai 500 da na matakai 2,000. Ga matakan masu zaman kansu da kansu, ya yi kuskure kusan 5% na lokaci, kamar yadda ya kamata. Jujjuya tsari yana lalata tsarin lokacin da tafiye-tafiyen ke ɗauke da shi, don haka yana gina ma’aunin kwatanci (reference) da ba ya bayyana tafiye-tafiye masu zaman kansu.

Ƙimomi ko matakai?

Binciken ya kuma lissafa ainihin matsakaici ga murabba’in covariance na nisa a iyaka (limiting squared distance covariance), 26/(75π) − 1/12, kusan 0.0270, wanda kwaikwayon suka sake samarwa a cikin kusan kuskuren ma’auni (standard error) ɗaya.

Marubutan sun fitar da darasi da taka-tsantsan. Tambayar ko hanyoyi biyu na bazata (processes) suna dogaro da juna ba ɗaya ba ce da tambayar ko wani biyu-biyun hanyoyi guda da aka lura da su suna da alaƙa cikin lokaci. Ɗaukar bambanci tsakanin ƙimomi masu bin juna yana dawo da matakai masu zaman kansu, amma a lokacin gwajin yana game da matakan ne, ba ƙimomin (levels) ba — wata tambaya ce daban. Ingantaccen gwaji ga bayanai masu yawo, marasa tsayayyen yanayi (non-stationary), dole ya saka dogaronsu cikin lokaci a cikin ma’aunin kwatancinsa, kuma gina irin wannan gwaji wata matsala ce daban da wannan takarda ba ta warware ba.