Hasashen Farashin Canjin Kuɗi EUR/USD ta Amfani da Hako Rubutu & Koyo Mai Zurfi: Hanyar PSO-LSTM

Teburin Abubuwan Ciki

1. Gabatarwa & Bayyani

Wannan bincike ya gabatar da wani sabon tsari na haɗaka don hasashen farashin canjin kuɗi EUR/USD, yana magance wani gibi mai mahimmanci a cikin ƙirar ƙididdiga na gargajiya ta hanyar haɗa bayanan rubutu masu inganci. Babban ƙirƙira ya ta'allaka ne akan haɗa ingantattun dabarun Sarrafa Harshe na Halitta (NLP)—musamman nazarin ra'ayi ta hanyar RoBERTa-Large da ƙirar jigo tare da Latent Dirichlet Allocation (LDA)—tare da injin hasashe mai zurfi wanda ya dogara da hanyoyin sadarwa na Ƙwaƙwalwar Ƙwaƙwalwa na Dogon Lokaci (LSTM). Ana ƙara inganta sigogin ƙirar ƙirar ta amfani da Ingantaccen Ƙwaƙwalwar Ƙungiyoyin Ƙwayoyin (PSO), ƙirƙirar ingantaccen tsarin hasashe mai dogaro da bayanai da ake kira PSO-LSTM.

Babban manufar binciken shine nuna cewa haɗa bayanan rubutu na ainihi, waɗanda ba a tsara su ba daga labarai da nazarin kuɗi yana haɓaka daidaiton hasashe sosai fiye da ƙirar da suka dogara kawai akan bayanan farashin da suka gabata. Ta yin haka, yana ɗaukar ra'ayin kasuwa da abubuwan motsa jigo waɗanda galibi suke gabatar da motsin kuɗi.

2. Hanyoyi & Tsarin Aiki

Hanyar da aka gabatar tana bin wani tsari mai tsari daga tarin bayanai daga tushe daban-daban har zuwa hasashe na ƙarshe.

2.1 Tattara Bayanai & Shirya Su

Bayanai na Ƙididdiga: An tattara tarihin farashin canjin kuɗi na yau da kullun na EUR/USD, gami da buɗewa, mafi girma, mafi ƙanƙanta, rufewa, da ƙarar ciniki. An samo alamomin fasaha (misali, matsakaicin motsi, RSI) a matsayin siffofi.

Bayanai na Rubutu masu Inganci: An tattara tarin labaran kuɗi da rahotannin nazarin kasuwa da suka shafi tattalin arzikin Tarayyar Turai da Amurka daga tushe masu mutunci. An tsaftace rubutun, an raba shi zuwa alamomi, kuma an shirya shi don nazarin NLP.

2.2 Hako Rubutu & Ƙirar Siffofi

Nazarin Ra'ayi: An daidaita ƙirar RoBERTa-Large da aka riga aka horar da ita akan bayanan ra'ayi na kuɗi don rarraba ra'ayin kowane labarin (tabbatacce, mara kyau, tsaka tsaki) kuma ta fitar da ci gaba da maki na ra'ayi. Wannan yana ba da ma'auni na ƙididdiga na yanayin kasuwa.

Ƙirar Jigo: An yi amfani da Latent Dirichlet Allocation (LDA) akan tarin rubutu don gano jigogi na ɓoye (misali, "Manufofin ECB," "Haɓakar Farashin Kayayyaki na Amurka," "Haɗarin Siyasa"). Rarraba jigo a kowace takarda da mahimman kalmomin jigo sun zama ƙarin siffofi, suna ɗaukar mahallin jigo na labarai.

Ƙarshen siffar siffa don kowane lokaci $t$ haɗaɗɗu ne: $\mathbf{X}_t = [\mathbf{P}_t, S_t, \mathbf{T}_t]$, inda $\mathbf{P}_t$ siffofi ne na ƙididdiga/fasaha, $S_t$ makin ra'ayi ne, kuma $\mathbf{T}_t$ siffar rarraba jigo ce.

2.3 PSO-LSTM Model Architecture

Ƙirar hasashe ita ce hanyar sadarwa ta LSTM, wacce aka zaɓa saboda ikonta na ƙirar dogon lokaci na dogaro a cikin bayanai masu tsari. Aikin tantanin halitta na LSTM a lokacin $t$ ana iya taƙaita shi da:

$\begin{aligned} \mathbf{f}_t &= \sigma(\mathbf{W}_f \cdot [\mathbf{h}_{t-1}, \mathbf{x}_t] + \mathbf{b}_f) \\ \mathbf{i}_t &= \sigma(\mathbf{W}_i \cdot [\mathbf{h}_{t-1}, \mathbf{x}_t] + \mathbf{b}_i) \\ \tilde{\mathbf{C}}_t &= \tanh(\mathbf{W}_C \cdot [\mathbf{h}_{t-1}, \mathbf{x}_t] + \mathbf{b}_C) \\ \mathbf{C}_t &= \mathbf{f}_t * \mathbf{C}_{t-1} + \mathbf{i}_t * \tilde{\mathbf{C}}_t \\ \mathbf{o}_t &= \sigma(\mathbf{W}_o \cdot [\mathbf{h}_{t-1}, \mathbf{x}_t] + \mathbf{b}_o) \\ \mathbf{h}_t &= \mathbf{o}_t * \tanh(\mathbf{C}_t) \end{aligned}$

Inda $\mathbf{x}_t$ shine siffar siffar shigarwa $\mathbf{X}_t$, $\mathbf{h}_t$ shine yanayin ɓoye, $\mathbf{C}_t$ shine yanayin tantanin halitta, kuma $\sigma$ shine aikin sigmoid.

An yi amfani da Ingantaccen Ƙwaƙwalwar Ƙungiyoyin Ƙwayoyin (PSO) don inganta mahimman sigogin LSTM (misali, adadin yadudduka, raka'a ɓoye, ƙimar koyo, ƙimar sauke). PSO tana binciken sararin sigogi ta hanyar kwaikwayon halayen zamantakewa na garken tsuntsaye, tana haɓaka mafita masu yuwuwa (barbashi) bisa ga nasu da mafi kyawun wuraren da aka sani na garken. Wannan yana sarrafawa kuma yana haɓaka tsarin daidaitawa idan aka kwatanta da bincike na hannu ko grid.

3. Sakamakon Gwaji & Bincike

3.1 Kwatanta da Ƙirar Ƙira na Ƙira

An kimanta ƙirar PSO-LSTM da wasu ƙa'idodin da aka kafa: Na'urar Tallafawa Vector (SVM), Regression na Tallafawa Vector (SVR), ARIMA, da GARCH. An auna aikin ta amfani da ma'auni na yau da kullun: Matsakaicin Kuskuren Cikakke (MAE), Tushen Matsakaicin Kuskuren Square (RMSE), da Matsakaicin Kuskuren Kashi na Cikakke (MAPE).

Bayanin Chati (Tunani): Chatin sandar da ke da taken "Kwatanta Aikin Hasashe (RMSE)" zai nuna sandar PSO-LSTM gajarta sosai (ƙananan kuskure) fiye da duk ƙirar ƙira. Chatin layi da ke rufe ainihin vs. farashin EUR/USD da aka hasashe zai nuna layin hasashen PSO-LSTM yana bin ainihin motsi sosai, yayin da layukan wasu ƙirar suka nuna babban karkata, musamman a lokutan sauyi da suka zo daidai da manyan abubuwan labarai.

Babban Bincike: Ƙirar PSO-LSTM ta ci gaba da fiye da duk ƙirar ƙira a cikin duk ma'auni na kuskure, yana nuna mafi girman ikon hasashe na haɗakar hanyar rubutu-ƙididdiga.

3.2 Binciken Cire Sassa (Ablation Study)

Don ware gudunmawar kowane ɓangaren bayanai, an gudanar da binciken cire sassa:

• Ƙirar A: LSTM tare da siffofi na ƙididdiga kawai (tushe).
• Ƙirar B: LSTM tare da ƙididdiga + siffofi na ra'ayi.
• Ƙirar C: LSTM tare da ƙididdiga + siffofi na jigo.
• Ƙirar D (Cikakke): PSO-LSTM tare da duk siffofi (ƙididdiga + ra'ayi + jigo).

Sakamako: Ƙirar D (Cikakke) ta sami mafi ƙarancin kuskure. Duka Ƙirar B da Ƙirar C sun yi kyau fiye da tushen Ƙirar A, suna tabbatar da cewa duka bayanan ra'ayi da jigo suna ƙara ƙima. Ribar aikin daga ƙara jigo ta ɗan fi na ƙara ra'ayi kawai a cikin wannan binciken, yana nuna cewa mahallin jigo sigina ne mai ƙarfi.

4. Zurfin Binciken Fasaha

4.1 Tsarin Lissafi

Babban matsalar hasashe an tsara shi azaman hasashen farashin canjin kuɗi na lokaci na gaba $y_{t+1}$ idan aka ba da jerin siffofin siffa na baya: $\hat{y}_{t+1} = f(\mathbf{X}_{t-n:t}; \mathbf{\Theta})$, inda $f$ shine ƙirar PSO-LSTM wanda $\mathbf{\Theta}$ ya ƙayyade, kuma $\mathbf{X}_{t-n:t}$ shine taga siffa mai tsayi $n$.

Algorithm ɗin PSO yana inganta sigogi $\mathbf{\Phi}$ (wani yanki na $\mathbf{\Theta}$) ta hanyar rage kuskuren hasashe akan saiti na tabbatarwa. Kowane barbashi $i$ yana da matsayi $\mathbf{\Phi}_i$ da sauri $\mathbf{V}_i$. Ƙididdigar sabuntawa su ne:

$\begin{aligned} \mathbf{V}_i^{k+1} &= \omega \mathbf{V}_i^k + c_1 r_1 (\mathbf{P}_{best,i} - \mathbf{\Phi}_i^k) + c_2 r_2 (\mathbf{G}_{best} - \mathbf{\Phi}_i^k) \\ \mathbf{\Phi}_i^{k+1} &= \mathbf{\Phi}_i^k + \mathbf{V}_i^{k+1} \end{aligned}$

inda $\omega$ shine rashin motsi, $c_1, c_2$ su ne ƙididdigar hanzari, $r_1, r_2$ lambobi ne na bazuwar, $\mathbf{P}_{best,i}$ shine mafi kyawun matsayin barbashi, kuma $\mathbf{G}_{best}$ shine mafi kyawun matsayi na duniya na garken.

4.2 Misalin Tsarin Bincike

Yanayi: Hasashen motsin EUR/USD don ranar ciniki ta gaba.

Mataki na 1 - Ciro Bayanai: Tsarin yana shan farashin rufewa, yana lissafin SMA na kwanaki 10, RSI (ƙididdiga). A lokaci guda, yana ciro sabbin kanun labarai 50 daga APIs na kuɗi da aka riga aka ayyana.

Mataki na 2 - Sarrafa Rubutu:

• Bututun Ra'ayi: Ana ciyar da kanun labarai cikin ƙirar RoBERTa-Large da aka daidaita. Fitowa: Matsakaicin makin ra'ayi na yau da kullun = -0.65 (mara kyau a matsakaici).
• Bututun Jigo: Ana sarrafa kanun labarai ta hanyar ƙirar LDA da aka horar. Fitowa: Babban jigo = "Manufofin Kuɗi" (nauyin 60%), tare da manyan kalmomi: "ECB," "lagarde," "farashin ruwa," "hawkish."

Mataki na 3 - Ƙirar Siffar Siffa: Haɗa: `[Close_Price=1.0850, SMA_10=1.0820, RSI=45, Sentiment_Score=-0.65, Topic_Weight_MonetaryPolicy=0.60, ...]`.

Mataki na 4 - Hasashe: Ana ciyar da siffar siffa cikin ƙirar PSO-LSTM da aka horar. Ƙirar, bayan ta koyi alamu kamar "ra'ayi mara kyau + jigon 'hawkish ECB' sau da yawa yana gabatar da ƙarfafa Yuro," tana fitar da hasashen dawowa.

Mataki na 5 - Fitowa: Ƙirar ta yi hasashen haɓakar +0.3% a cikin EUR/USD don washegari.

5. Aikace-aikace na Gaba & Jagorori

Tsarin yana da iya faɗaɗawa sosai. Jagororin gaba sun haɗa da:

• Hasashe na Ainihi: Ƙaddamar da ƙirar a cikin tsarin gudana don hasashe na cikin rana ta amfani da ciyarwar labarai mai yawan mita da bayanan tikiti.
• Kayan Aiki Da Yawa & Nau'ikan Kuɗi Masu Keta: Yin amfani da hanyar guda ɗaya don hasashen wasu manyan nau'ikan FX (misali, GBP/USD, USD/JPY) ko ma farashin kuɗin dijital, waɗanda aka san su da ra'ayi.
• Haɗaɗɗun Bayanai na Madadin: Haɗa sigina daga kafofin watsa labarun zamantakewa (misali, ra'ayi na Twitter/X), rubutun jawabin bankunan tsakiya da aka bincika tare da ingantattun LLMs, ko bayanan hotunan tauraron dan adam don ayyukan tattalin arziki, bin yanayin da aka gani a cikin binciken asusun shinge.
• Ingantaccen Tsarin Gine-gine: Maye gurbin LSTM na yau da kullun tare da bambance-bambancen da suka fi ƙware kamar ƙirar tushen Transformer (misali, Na'urori na Haɗaɗɗun Lokaci) ko ƙirar haɗakar CNN-LSTM don ɗaukar duka alamu na sarari a cikin siffofi da dogaro na lokaci.
• AI Mai Bayyanawa (XAI): Haɗa kayan aiki kamar SHAP ko LIME don fassara yanke shawarar ƙirar, gano waɗanne takamaiman jigogin labarai ko sauye-sauyen ra'ayi suka fi tasiri ga wani hasashe, mai mahimmanci don samun amincewa a aikace-aikacen kuɗi.

6. Nassoshi

1. Hochreiter, S., & Schmidhuber, J. (1997). Long short-term memory. Neural computation.
2. Kennedy, J., & Eberhart, R. (1995). Particle swarm optimization. Proceedings of ICNN'95.
3. Liu, Y., et al. (2019). RoBERTa: A Robustly Optimized BERT Pretraining Approach. arXiv preprint arXiv:1907.11692.
4. Blei, D. M., Ng, A. Y., & Jordan, M. I. (2003). Latent Dirichlet allocation. Journal of machine Learning research.
5. Box, G. E. P., Jenkins, G. M., & Reinsel, G. C. (2008). Time Series Analysis: Forecasting and Control. Wiley.
6. Vaswani, A., et al. (2017). Attention is All You Need. Advances in Neural Information Processing Systems.
7. Investopedia. (2023). Kasuwar Canjin Kuɗi na Waje (Forex). An samo daga investopedia.com.
8. Bankin Turai na Tsakiya & Bayanan Tattalin Arziki na Tarayya (FRED) – a matsayin wakilai na tushen bayanai na asali.

7. Bincike Mai Zurfi na Manazarcin