1. Gabatarwa
Wannan binciken ya magance kalubalen gina fayil zuba jari mafi kyau na kaya biyu ta amfani da fasahohin injin koyo. Binciken ya mayar da hankali kan fayil da ya ƙunshi ma'aunin hannun jari na S&P 500 da haɗin kuɗin USD/GBP. Babban manufar ita ce yin amfani da bayanan tattalin arziki a mitar wata-wata da na kwata-kwata don hasashen dawowar waɗannan kadarorin ta amfani da hanyoyin tarukan bishiyoyi—musamman Dajin Bazuwar da XGBoost. Ana amfani da waɗannan hasashe a matsayin shigarwar da ake tsammanin dawowa don ingantaccen ka'idar fayil (MPT). Manufa ta biyu ita ce gano waɗanne masu canjin tattalin arziki ke da tasiri mai mahimmanci akan hasashe. Binciken ya yi nazari mai zurfi kan ko fayil da aka gina akan bayanan da aka ƙididdige ta hanyar ML ya bambanta da gaske da wanda aka gina ta amfani da matsakaicin tarihi mai sauƙi.
2. Hanyar Aiki & Bayanai
2.1 Tattara Bayanai & Shirye-shiryen Farko
Nazarin yana amfani da bayanan lokaci-lokaci don ma'aunin S&P 500 da farashin musayar USD/GBP. Ana tattara jerin alamomin tattalin arziki a matsayin fasalolin hasashe masu yuwuwa, waɗanda ƙila su haɗa da masu canji kamar ƙimar riba, ma'aunun hauhawar farashin kayayyaki, alkaluman samar da masana'antu, da ƙimar rashin aikin yi, waɗanda aka samo daga cibiyoyin bayanai kamar FRED. An raba bayanan zuwa saitin horo, tabbatarwa, da gwaji, tare da kulawa sosai don guje wa son zuciya na gaba. Ana daidaita fasalolin ko daidaita su kamar yadda samfurori suka buƙata.
2.2 Samfuran Tarukan Bishiyoyi: Dajin Bazuwar & XGBoost
An yi amfani da algorithms guda biyu na haɗakar koyo don hasashen lokaci-lokaci:
- Dajin Bazuwar: Haɗakar bishiyoyin yanke shawara da aka horar da su akan samfuran bootstrapped na bayanai tare da zaɓin fasalin bazuwar, yana rage yawan dacewa da kuma samar da hasashe masu ƙarfi.
- XGBoost (Haɓaka Gradient Mai Tsanani): Tsarin haɓaka gradient mai daidaitawa, wanda aka sani da saurinsa da aiki. Yana gina bishiyoyi a jere don gyara kurakuran na baya, sau da yawa yana ba da sakamako na zamani akan bayanan da aka tsara.
An zaɓi waɗannan samfuran saboda ikonsu na sarrafa alaƙar da ba ta layi ba da kuma mu'amala masu rikitarwa tsakanin masu canjin tattalin arziki ba tare da ƙa'idodin ƙididdiga masu tsauri ba.
2.3 Tsarin Gina Fayil
Dawowar da aka hasashe daga samfuran ML suna aiki a matsayin vector da ake tsammanin dawowa $\mu$ a cikin tsarin ingantaccen ma'aunin ma'anar bambance-bambancen Markowitz. Ana ƙayyade ma'aunin nauyin fayil $w$ na kadarorin biyu ta hanyar warware matsalar ingantawa wacce ke haɓaka ma'aunin Sharpe ko rage bambance-bambancen don dawowar da aka yi niyya. Ana ƙididdige matrix covariance $\Sigma$ yawanci daga dawowar tarihi. Ana kwatanta aikin "fayil na tushen ML" da fayil na ma'auni da aka gina ta amfani da matsakaicin dawowar tarihi.
3. Sakamakon Gwaji & Nazari
3.1 Aikin Hasashe
Samfuran tarukan bishiyoyi sun nuna ikon ƙididdiga mai mahimmanci na hasashen motsin alkibla da, zuwa wani ƙaramin mataki, girman dawowar duka S&P 500 da USD/GBP. An ba da rahoton ma'auni na kimantawa kamar Kuskuren Matsakaicin Cikakke (MAE), Tushen Kuskuren Matsakaicin Murabba'i (RMSE), da daidaiton alkibla. XGBoost sau da yawa yana nuna ɗan gaba a kan Dajin Bazuwar dangane da daidaiton hasashe, musamman akan bayanan kwata-kwata, mai yiwuwa saboda ingantaccen tsarin haɓakawa da daidaitawa.
3.2 Kwatancen Aikin Fayil
Bayanin Ginshiƙi: Zane mai kwatancen zai nuna jimillar dawowar fayiloli uku a cikin lokacin gwaji na waje: 1) Fayil mafi kyau na tushen hasashen ML, 2) Fayil mafi kyau na tushen matsakaicin tarihi, da 3) Ma'auni mai daidaitaccen ma'auni.
Sakamakon ya nuna cewa fayil da aka gina ta amfani da hasashen ML ya sami mafi kyawun bayanin dawowar da aka daidaita haɗari (ma'aunin Sharpe mafi girma) idan aka kwatanta da fayil na tushen matsakaicin tarihi. Ma'aunin rabon kadarori tsakanin S&P 500 da USD/GBP suma sun bambanta da ma'ana, yana nuna samfuran ML sun kama dawowar da ake tsammani masu canzawa lokaci-lokaci waɗanda matsakaicin tarihi mai sauƙi ba zai iya ba.
3.3 Nazarin Muhimmancin Fasali
Duka Dajin Bazuwar da XGBoost suna ba da maki mahimmanci na fasali na asali. Nazarin ya bayyana cewa ga S&P 500, alamomin jagora kamar shimfiɗar lokaci, halin masu amfani, da rashin kwanciyar hankali na kasuwar hannun jari na baya suna cikin manyan masu hasashe. Ga USD/GBP, bambance-bambancen ƙimar riba, bayanan ma'aunin ciniki, da manyan motsin ma'aunin dala sun fi tasiri. Wannan fahimtar yana da mahimmanci don fassarar tattalin arziki da sauƙaƙe samfuri.
4. Muhimman Fahimta & Tattaunawa
Fahimta ta Asali
Mafi ƙarfin hujjar takardar ba ita ce ML zai iya doke kasuwa ba—amma cewa ko da ƙananan ingantattun hasashe ta hanyar tarukan bishiyoyi na iya canza lissafin iyaka mai inganci don sauƙi na fayil na kaya biyu. Wannan yana ƙalubalantar ra'ayin rabon "saita-da-manta" na dogon lokaci ga masu zuba jari a cikin gaurayawan da ba na hannun jari/bashi ba.
Tsarin Hankali
Hankalin binciken yana da inganci: 1) Yi amfani da ML mai ƙarfi, mara ƙididdiga (RF/XGBoost) don narkar da bayanan macro zuwa hasashen dawowa, tare da kaucewa tarko na samfurin layi. 2) Cusa waɗannan hasashe a cikin injin Markowitz na gargajiya. 3) Tabbatar cewa fayil ɗin da aka fitar ya bambanta da ma'auni na tarihi. Gudu daga masu tuka macro zuwa hasashen kadarori zuwa ma'aunin nauyin fayil yana bayyana kuma ana iya maimaitawa.
Ƙarfi & Aibobi
Ƙarfi: Mayar da hankali mai aiki akan lamarin kaya biyu mai sauƙi yana haɓaka bayyanawa. Yin amfani da samfuran bishiyoyi yana ba da rashin layi na asali da mahimmanci na fasali, yana ƙara fassarar tattalin arziki sau da yawa ba ta nan a cikin takardun kuɗi na zurfin koyo. Kwatanta da ma'auni na matsakaicin tarihi yana da adalci kuma yana da alaƙa.
Aibobi: Giwa a cikin ɗaki shine ƙididdigar covariance. Binciken yana amfani da covariance na tarihi, wanda aka sani da rashin kwanciyar hankali. Tsarin covariance da aka hasashen ML zai iya zama mataki na gaba na hankali amma ba ya nan. Sauƙaƙa kaya biyu, yayin da yake da ƙarfi don bayyanawa, yana iyakance fa'idodin bambance-bambancen da ML zai iya buɗewa a cikin mahallin kadarori da yawa. Ba a magance farashin ciniki da yuwuwar aikin sake daidaita wata-wata/kwata-kwata bisa waɗannan sigina ba.
Fahimta Mai Aiki
Ga masu aiki: Kada ku yi watsi da hanyoyin haɗakarwa masu sauƙi kamar XGBoost don hasashen dawowa; suna iya zama masu ƙarfi da fassara fiye da jijiyoyin jijiyoyi don bayanan macro/kuɗi da aka tsara. Manyan masu tuka macro da aka gano (misali, shimfiɗar lokaci don hannun jari, bambance-bambancen ƙimar riba don FX) yakamata su kasance a kan tunanin masu nazari waɗanda ke sa ido kan waɗannan nau'ikan kadarorin. Wannan hanya ta fi dacewa ga masu zuba jari na hukumomi ko ƙwararrun mutane waɗanda za su iya aiwatar da tsarin da sake daidaita irin wannan dabarar, ba ga 'yan kasuwa na dillalai waɗanda ke neman alpha na ɗan gajeren lokaci ba.
5. Cikakkun Bayanai na Fasaha & Tsarin Lissafi
Jigon ingantaccen fayil shine samfurin bambance-bambancen ma'anar Markowitz. Manufar ita ce nemo vector nauyi $w$ wanda ke warware ɗaya daga cikin matsaloli biyu:
Matsakaicin Ma'aunin Sharpe:
$\max_{w} \frac{w^T \mu}{\sqrt{w^T \Sigma w}}$
ƙarƙashin sharuɗɗa $\sum_i w_i = 1$, da yuwuwar $w_i \ge 0$ don rashin sayar da gajere.
Mafi ƙarancin Bambance-bambance don Dawowar Manufa $R_p$:
$\min_{w} w^T \Sigma w$
ƙarƙashin sharuɗɗa $w^T \mu = R_p$ da $\sum_i w_i = 1$.
Inda $\mu$ shine vector na dawowar da ake tsammani (wanda RF/XGBoost ya hasashe) kuma $\Sigma$ shine matrix covariance na dawowa. Samfuran tarukan bishiyoyi da kansu suna aiki ta hanyar ƙirƙirar saitin bishiyoyi $M$ (don Dajin Bazuwar) ko bishiyoyi da aka gina a jere (don XGBoost) waɗanda ke tsara fasalolin shigarwa $x$ zuwa hasashen dawowa $\hat{y}$. Ga Dajin Bazuwar, hasashen shine matsakaici: $\hat{y} = \frac{1}{M} \sum_{m=1}^{M} T_m(x)$. Hasashen XGBoost shine samfuri mai ƙari: $\hat{y} = \sum_{k=1}^{K} f_k(x)$, inda kowane $f_k$ bishiya ce daga sararin aiki $\mathcal{F}$, kuma an horar da samfurin ta hanyar rage manufa mai daidaitawa: $\mathcal{L}(\phi) = \sum_i l(\hat{y}_i, y_i) + \sum_k \Omega(f_k)$, tare da $\Omega(f) = \gamma T + \frac{1}{2}\lambda ||w||^2$ yana sarrafa rikitarwa.
6. Tsarin Nazari: Misalin Hali
Yanayi: Asusun zuba jari yana son raba tsakanin hannun jari na Amurka (wanda ETF na SPY ke wakilta) da farashin musayar GBP/USD (wanda matsayin forex ke wakilta) don kwata mai zuwa.
Mataki 1 - Shirye-shiryen Bayanai: Tattara bayanan watanni 10 da suka gabata na dawowar SPY, dawowar GBP/USD, da masu canjin tattalin arziki 20 (misali, CPI na Amurka, CPI na Burtaniya, Ƙimar Asusun Fed, Ƙimar BoE, shimfiɗar yawan amfanin 10Y na Amurka-Burtaniya, VIX, da sauransu). Maɓallin maɓalli shine dawowar lokaci na gaba. An ajiye shekaru 2 na baya-bayan nan a matsayin saitin gwaji.
Mataki 2 - Horar da Samfuri & Hasashe: Horar da samfurin XGBoost akan bayanan horo don hasashen dawowar SPY da wani samfuri daban don dawowar GBP/USD. Yi amfani da daidaita hyperparameter (ta hanyar gwaji-gwaji) don sigogi kamar `max_depth`, `learning_rate`, da `n_estimators`. Samar da hasashe gaba ɗaya don lokacin gwaji.
Mataki 3 - Ingantaccen Fayil: Ga kowane wata a cikin saitin gwaji, yi amfani da hasashen XGBoost a matsayin $\mu$ da dawowar tarihi na shekaru 3 da suka gabata don ƙididdige matrix covariance $\Sigma$. Warware ma'aunin nauyin fayil na tangency (ma'aunin Sharpe mafi girma).
Mataki 4 - Gwajin Baya & Kimantawa: Ƙididdige jimillar dawowa, rashin kwanciyar hankali, da ma'aunin Sharpe na fayil na tushen ML da aka sake daidaita. Kwatanta shi da fayil mai tsayayye 60/40 da fayil da ke amfani da matsakaicin dawowar tarihi don $\mu$.
7. Aikace-aikacen Gaba & Hanyoyin Bincike
- Fayiloli na Kaya Da Yawa: Tsawaita tsarin zuwa sararin samaniya mai faɗi na kadarori (lamuni, kayayyaki, hannun jari na ƙasashen waje) don gwada ikon bambance-bambancen na gaske na ML.
- Ƙididdigar Covariance Mai Ƙarfi: Haɗa fasahohin ML (misali, Graphical LASSO, RNNs) don hasashen matrix covariance $\Sigma$ tare da dawowa, matsawa bayan ƙididdigar tarihi.
- Haɗa Bayanan Madadin: Haɓaka saitin fasali tare da bayanan halayen daga labarai/kafofin watsa labarai, bayanan sarkar wadata, ko hotunan tauraron dan adam, kamar yadda aka bincika a cikin bincike kamar "Tasirin Labarai akan Rashin Kwanciyar Hankali" (Tetlock, 2007).
- Koyo Kan Layi & Daidaitawa: Aiwatar da nau'ikan tarukan bishiyoyi kan layi waɗanda za su iya daidaitawa da canje-canjen yanayin kasuwa a ainihin lokaci, ra'ayi daidai da kalubalen "koyo na ci gaba" a cikin AI.
- Haɗin AI Mai Bayyanawa (XAI): Yin amfani da ƙimar SHAP (SHapley Additive exPlanations) tare da mahimmanci na fasali don samar da zurfin bayani, matakin misali don dalilin da ya sa aka yi wani hasashe, mai mahimmanci ga amincewar masu ruwa da tsaki a cikin kuɗi.
- Haɗin Factor: Haɗa hasashen ML tare da samfuran factor na gargajiya (misali, abubuwan Fama-French) don ƙirƙirar ƙididdiga na tsammanin dawowar hybrid.
8. Nassoshi
- Ahmed, N. K., Atiya, A. F., Gayar, N. E., & El-Shishiny, H. (2010). An empirical comparison of machine learning models for time series forecasting. Econometric Reviews, 29(5-6), 594-621.
- Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5-32.
- Chen, T., & Guestrin, C. (2016). XGBoost: A scalable tree boosting system. Proceedings of the 22nd ACM SIGKDD International Conference on Knowledge Discovery and Data Mining, 785-794.
- Markowitz, H. (1952). Portfolio selection. The Journal of Finance, 7(1), 77-91.
- Pham, H. (2025). [Aikin da ya dace akan dabarun hannun jari/lamuni da aka ambata a cikin PDF].
- Ţiţan, A. G. (2015). The efficient market hypothesis: Review of specialized literature and empirical research. Procedia Economics and Finance, 32, 442-449.
- Zhu, J. Y., Park, T., Isola, P., & Efros, A. A. (2017). Unpaired image-to-image translation using cycle-consistent adversarial networks. Proceedings of the IEEE International Conference on Computer Vision, 2223-2232. (An ambata a matsayin misali na takarda na tsarin ML na asali don tunani).