Tsinkayar da Hasashen Rashin Kwanciyar Hankali na Musanya Kudade a Cikin Yankin Lokaci da Mita

Tsarin Abubuwan Ciki

1. Gabatarwa & Bayyani

Wannan takarda ta gabatar da wata sabuwar hanya don tsarawa da tsinkayar rashin kwanciyar hankali na kuɗi, musamman ga farashin musanya, ta hanyar haɗa binciken bayanai na girma-mita tare da dabarun rarrabuwa na lokaci-mita. Babban ƙirƙira ya ta'allaka ne a cikin haɓaka tsarin Realized GARCH tare da ma'aunin rashin kwanciyar hankali da aka gane da aka rarraba ta hanyar wavelet da kuma na'urar kima ta musamman don tsalle. Wannan yana ba da damar ƙirar ta rarraba rashin kwanciyar hankali zuwa sassa masu dacewa da madaidaitan lokutan saka hannun jari (matsakaicin lokaci) kuma a ɗauki tasirin tsallen farashi na katsewa daban. Binciken ya samo asali ne daga yanayin ɗimbin ɗimbin mahalarta kasuwa waɗanda ke aiki a kan madaidaitan lokuta daban-daban, tun daga 'yan kasuwa na girma-mita zuwa masu saka hannun jari na dogon lokaci.

Marubutan sun nuna cewa ƙirar su da aka gabatar na "Jump-GARCH", waɗanda aka ƙididdige su ta hanyar Mafi Girman Yiwuwar (MLE) da kuma Tsarin Maki na Autoregressive Gama gari (GAS), suna ba da tsinkaye mafi girma a ƙididdiga idan aka kwatanta da na al'ada na GARCH da shahararrun ƙirar rashin kwanciyar hankali da aka gane. Binciken yana amfani da bayanan gaba na musanya kuɗin waje waɗanda suka haɗa da rikicin kuɗi na 2007-2008, yana ba da gwajin matsi mai ƙarfi ga hanyar.

2. Hanyoyin Nazari & Tsarin Fasaha

2.1 Tsarin Realized GARCH

Ƙirar Realized GARCH tana cike gibin tsakanin ƙirar GARCH na al'ada da bayanai na girma-mita ta hanyar haɗa ma'aunin rashin kwanciyar hankali da aka gane $RV_t$ kai tsaye cikin ma'auni na rashin kwanciyar hankali. Tsarin asali ya ƙunshi ma'auni na dawowa, ma'auni na GARCH don rashin kwanciyar hankali na ɓoye, da ma'aunin ma'auni wanda ke haɗa rashin kwanciyar hankali na ɓoye zuwa ma'aunin da aka gane.

2.2 Rarrabuwa ta Tushen Wavelet a Matsakaici Daban-daban

Don ɗaukar yanayin rashin kwanciyar hankali mai matakai daban-daban, marubutan sun yi amfani da canjin wavelet. Wannan kayan aikin lissafi yana rarraba jerin rashin kwanciyar hankali da aka gane zuwa sassa masu kai tsaye waɗanda ke wakiltar madaidaitan lokuta daban-daban (misali, motsi na cikin rana, na yau da kullun, na mako). Idan $RV_t$ shine rashin kwanciyar hankali da aka gane, rarrabuwarsa na wavelet za a iya wakilta shi kamar haka:

$RV_t = \sum_{j=1}^J D_{j,t} + S_{J,t}$

inda $D_{j,t}$ ke wakiltar ɓangaren rashin kwanciyar hankali ("cikakkun bayanai") a ma'auni $j$ (wanda ya dace da takamaiman rukunin mita), kuma $S_{J,t}$ shine ɓangaren santsi wanda ke ɗaukar yanayin dogon lokaci. Kowane $D_{j,t}$ yana kusantar ayyukan ciniki da kwararar bayanai a takamaiman madaidaicin lokacin saka hannun jari.

2.3 Gano Tsalle & Mai Kima na JTSRV

Wani ci gaba mai mahimmanci shine haɗa bambancin tsalle. Marubutan sun yi amfani da mai kima na Jump Two Scale Realized Volatility (JTSRV). Wannan mai kima yana raba jimlar bambancin quadratic zuwa bambancin haɗe-haɗe mai ci gaba (IV) da bambancin tsalle mai katsewa (JV):

$RV_t \approx IV_t + JV_t$

Wannan rabuwa yana da mahimmanci saboda tsalle da rashin kwanciyar hankali mai ci gaba sau da yawa suna da dagewa da kaddarorin tsinkaya daban-daban.

2.4 Kima: MLE da GAS

An ƙididdige ƙirar Jump-GARCH da aka gabatar ta amfani da hanyoyi biyu: 1) Ƙididdigar Mafi Girman Yiwuwar Quasi-Maximum Likelihood Estimation (QMLE), da kuma 2) tsarin Maki na Autoregressive Gama gari (GAS) wanda aka ƙaddamar da shi ta hanyar bayanai. Tsarin GAS, wanda Creal da sauransu (2013) suka gabatar, yana sabunta sigogi bisa ga maki na aikin yiwuwar, yana ba da yuwuwar ƙarfi da daidaitawa ga kuskuren ƙira.

3. Nazarin Aiki & Sakamako

3.1 Bayanai & Tsarin Gwaji

Binciken yana amfani da bayanai na girma-mita don gaba na FX (mai yiwuwa manyan nau'ikan biyu kamar EUR/USD). Lokacin samfurin ya haɗa da rikicin kuɗi na 2007-2009, yana ba da damar bincikin aikin ƙira a ƙarƙashin matsi mai tsanani. An kimanta tsinkaye don madaidaitan lokuta na gaba ɗaya da na lokuta da yawa.

3.2 Aikin Tsinkaya

An yi amfani da ƙirar da aka gabatar a matsayin ma'auni da ƙirar daidaitattu kamar GARCH(1,1) da HAR-RV. Kimantawa yana amfani da ayyukan asara na ƙididdiga (misali, MSE, QLIKE). An gabatar da sakamako mafi mahimmanci a cikin tebur mai kwatanta (wanda aka kwaikwayi a ƙasa):

Lura: Ƙimomin rabo ne na kwatancen da ma'aunin GARCH(1,1).

3.3 Babban Bincike & Fahimta

• Rabuwar Tsalle Ita Ce Maɓalli: Rarraba bambancin tsalle daga bambancin haɗe-haɗe yana ci gaba da inganta daidaiton tsinkaya.
• Rinjayen Girma-Mita: Madaidaicin lokaci mafi cike da bayanai don rashin kwanciyar hankali na gaba shine ɓangaren girma-mita (madaidaicin lokaci gajere) na rarrabuwar wavelet.
• Fifikon Ƙira: Sabbin ƙirar Jump-GARCH tare da rarrabuwar wavelet sun fi ƙirar GARCH na al'ada da na Realized GARCH na daidaitattu a ƙididdiga.
• Ƙarfin Jurewa Rikici: Ƙirar sun nuna aiki mai ƙarfi a lokacin rikicin kuɗi.

4. Babban Fahimta & Ra'ayi na Manazarta

Babban Fahimta: Wannan takarda tana isar da saƙo mai ƙarfi, amma ba a yaba da shi ba: rashin kwanciyar hankali ba tsari guda ɗaya ba ne amma tsari ne mai yawa. Ta hanyar ƙin ɗaukar kasuwa a matsayin abu guda ɗaya, mai daidaitaccen yanayi, a maimakon haka ta yin amfani da wavelet don rarraba shi zuwa madaidaitan lokutan saka hannun jari na asali, marubutan sun buɗe akwatin baƙin na motsin rashin kwanciyar hankali. Gano cewa ɓangarorin gajeren lokaci, girma-mita suna tafiyar da tsinkaye kalubale ne kai tsaye ga ƙirar da suka fi ɗaukar nauyin yanayin dogon lokaci kuma suna jaddada ƙaruwar rinjayen algorithmic da ciniki na girma-mita a cikin gano farashi da samuwar rashin kwanciyar hankali.

Kwararar Ma'ana: An gina hujja cikin kyau. Ya fara ne daga tabbataccen gaskiyar aiki na ɗimbin wakilan kasuwa (daga ƙirar HAR na Corsi). Sa'an nan kuma a tambaya cikin ma'ana: idan wakilai suna aiki akan madaidaitan lokuta daban-daban, shin ƙirar mu bai kamata su nuna hakan ba? Rarrabuwar wavelet ita ce amsar fasaha ta cikakke. Haɗin haɗarin tsalle na gaba—wani gaskiyar kasuwa mara Gaussian, mai katsewa—ya cika hoton. Kwararar daga fahimtar tattalin arziki (ɗimbin yawa) zuwa kayan aikin lissafi (wavelet) zuwa sakamakon aiki (ingantaccen tsinkaya) yana da gamsarwa.

Ƙarfi & Kurakurai: Babban ƙarfi shine nasarar haɗa ƙididdiga masu zurfi (Realized GARCH, wavelet, gano tsalle) cikin tsari mai ma'ana, mai nasara a aiki. Ya wuce kwatancen ƙira mai sauƙi don ba da fahimta ta gaske game da tushen tsinkaya. Amfani da tsarin GAS kuma yana da hangen nesa. Babban aibi, gama gari a cikin wannan wallafe-wallafen, shine jin "a cikin samfurin" na duba ƙarfi. Duk da yake lokacin rikici ya haɗa, gwajin gaske na waje akan bayanan da ba a gani gaba ɗaya ba (misali, faɗuwar COVID na 2020) zai fi gamsarwa. Bugu da ƙari, rikitaccen lissafi na ƙirar wavelet-GARCH-tsalle na iya iyakance aikace-aikacen sa na ainihi a wasu tsarin ciniki, wani matsala mai amfani da ba a magance ba.

Fahimta Mai Aiki: Ga masu ƙididdiga da masu kula da haɗari, wannan takarda tsari ne. Na farko, ka rarraba, sannan ka ƙira. Yin amfani da tace wavelet mai sauƙi akan jerin rashin kwanciyar hankalin ku kafin a ciyar da shi cikin ƙirar ML ko ƙididdiga da kuka fi so na iya haifar da riba nan take. Na biyu, ka ɗauki tsalle daban. Gina siginar keɓantacce don gano tsalle da ƙirar tasirinsa da kansa, kamar yadda aka yi tare da JTSRV, shine mafi kyawun aiki marar sasantawa ga kowane ƙirar rashin kwanciyar hankali mai mahimmanci bayan 2008. A ƙarshe, mayar da hankalin ku na tsinkaya akan Layer na girma-mita. Rarraba ƙarin bincike da albarkatun lissafi don fahimta da tsinkayar motsin rashin kwanciyar hankali na cikin rana, domin a nan ne mafi girman siginar tsinkaya ke nan.

5. Cikakkun Bayanai na Fasaha & Tsarin Lissafi

Za a iya taƙaita babban ƙirar Jump-GARCH tare da ɓangarorin wavelet kamar haka:

Ma'auni na Dawowa: $r_t = \sqrt{h_t} z_t$, inda $z_t \sim i.i.d.(0,1)$.

Ma'auni na GARCH: $h_t = \omega + \beta h_{t-1} + \gamma \xi_{t-1}$.

Ma'aunin Ma'auni (An Inganta):
$\log(RV_t) = \xi + \phi \log(h_t) + \tau_1 z_t + \tau_2 (z_t^2 - 1) + \sum_{j=1}^J \delta_j D_{j,t} + \lambda J_t + u_t$
inda $u_t \sim i.i.d.(0, \sigma_u^2)$. A nan, $D_{j,t}$ su ne ɓangarorin cikakkun bayanai na wavelet na $RV_t$, kuma $J_t$ shine ɓangaren tsalle mai mahimmanci wanda mai kima na JTSRV ya gano.

Ƙirar tana ƙididdige sigogi $\theta = (\omega, \beta, \gamma, \xi, \phi, \tau_1, \tau_2, \{\delta_j\}, \lambda)$ don ɗaukar motsi tsakanin rashin kwanciyar hankali na ɓoye, ma'auni da aka gane, tsalle, da ɓangarorin matakai daban-daban.

6. Tsarin Nazari: Misalin Lamari

Yanayi: Asusun shinge na ƙididdiga yana son inganta tsinkayar sa na yau da kullun na Ƙimar Haɗari (VaR) don littafin ciniki na EUR/USD.

Mataki 1 - Shirya Bayanai: Sami dawowar cikin rana na mintuna 5 don EUR/USD. Ƙididdige rashin kwanciyar hankali da aka gane na tushe (misali, RV) kuma a yi amfani da canjin wavelet (ta amfani da ɗakin karatu kamar PyWavelets a cikin Python) don rarraba shi zuwa ma'auni 3: D1 (motsi na awa 2-4), D2 (awa 4-8), D3 (awa 8-16). Daban, yi amfani da mai kima na JTSRV don cire jerin tsalle na yau da kullun $J_t$.

Mataki 2 - Ƙayyadaddun Ƙira & Kima: Ƙididdige ƙirar Jump-GARCH daga Sashe na 5, inda ma'aunin ma'auni ya haɗa da D1, D2, D3, da $J_t$ a matsayin masu canji na waje. Kwatanta yiwuwar log da ma'auni na bayanai tare da daidaitaccen ƙirar Realized GARCH.

Mataki 3 - Tsinkaya & Aikace-aikace: Samar da tsinkayar rashin kwanciyar hankali na gaba ɗaya $\hat{h}_{t+1}$ daga ƙirar da aka ƙididdige. Yi amfani da wannan tsinkaya don ƙididdige VaR (misali, $VaR_{t+1}^{\alpha} = -\Phi^{-1}(\alpha) \sqrt{\hat{h}_{t+1}}$). Gwada tsinkayar VaR da ainihin P&L don kimanta daidaiton ɗaukar hoto.

Sakamako da ake tsammani: Tsinkayar VaR daga ƙirar Jump-GARCH tare da wavelet ya kamata ya nuna ingantaccen ɗaukar hoto (ƴan keɓancewa) kuma ya zama ƙasa da saukin ƙima ƙarancin haɗari bayan kwanaki masu tsalle mai girma ko takamaiman tsarin rashin kwanciyar hankali na cikin rana.

7. Aikace-aikace na Gaba & Hanyoyin Bincike

• Haɗa Kayan Koyon Injina: ɓangarorin wavelet $D_{j,t}$ da jerin tsalle $J_t$ na iya zama siffofi masu cike da bayanai don ƙirar koyon injina (misali, LSTM, Gradient Boosting) don tsinkayar rashin kwanciyar hankali, suna motsawa bayan tsarin GARCH na layi/paramita.
• Zubar da Rashin Kwanciyar Hankali na Kayan Dukiya: Yi amfani da rarrabuwa mai matakai daban-daban don nazarin yadda rashin kwanciyar hankali ke watsawa tsakanin nau'ikan kadari (misali, daga hannun jari zuwa FX) a madaidaitan lokuta daban-daban. Shin faɗuwar kasuwar hannun jari tana watsawa ta ɓangarorin rashin kwanciyar hankali na gajeren lokaci ko na dogon lokaci?
• Siginonin Ciniki na Ainihi: Haɓaka dabarun ciniki waɗanda ke amfani da bambanci tsakanin ɓangarorin rashin kwanciyar hankali na gajeren lokaci da na dogon lokaci a matsayin siginar komawa ga matsakaici ko motsi.
• Binciken Bankin Tsakiya & Manufofi: Yi amfani da tsarin don nazarin tasirin sanarwar manufofin kuɗi akan rashin kwanciyar hankali na FX, tare da bambanta tsakanin "tsallen labarai" na girma-mita nan take da ɗaukar bayanai na dogon lokaci.
• Ƙaddamarwa zuwa Kuɗin Sirri: Gwada ƙirar akan kasuwannin kuɗin sirri na 24/7, waɗanda ke da halayen tsalle mai tsanani da halayen masu saka hannun jari masu matakai daban-daban, tun daga ƙwararrun mutum-mutumi na algorithmic zuwa "HODLers" na dogon lokaci.

8. Nassoshi

1. Barunik, J., Krehlik, T., & Vacha, L. (2015). Modeling and forecasting exchange rate volatility in time-frequency domain. Preprint, arXiv:1204.1452v4.
2. Corsi, F. (2009). A simple approximate long-memory model of realized volatility. Journal of Financial Econometrics, 7(2), 174-196.
3. Hansen, P. R., & Lunde, A. (2005). A forecast comparison of volatility models: does anything beat a GARCH(1,1)? Journal of Applied Econometrics, 20(7), 873-889.
4. Creal, D., Koopman, S. J., & Lucas, A. (2013). Generalized autoregressive score models with applications. Journal of Applied Econometrics, 28(5), 777-795.
5. Gençay, R., Selçuk, F., & Whitcher, B. (2005). Multiscale systematic risk. Journal of International Money and Finance, 24(1), 55-70.
6. McAleer, M., & Medeiros, M. C. (2008). A multiple regime smooth transition heterogeneous autoregressive model for long memory and asymmetries. Journal of Econometrics, 147(1), 104-119.
7. Andersen, T. G., & Bollerslev, T. (1998). Answering the skeptics: Yes, standard volatility models do provide accurate forecasts. International Economic Review, 39(4), 885-905.