Loading [Contrib]/a11y/accessibility-menu.js
Research article

Integrating LSTM with Fama-French six factor model for predicting portfolio returns: Evidence from Shenzhen stock market China

  • The study aimed to examine the effectiveness of long short-term memory (LSTM) model in predicting portfolio returns employing Fama and French's six-factor model. Monthly data were collected for A-share prices for firms listed on the Shenzhen Stock Exchange, China, for the period extended over 25 years (1997–2022). Portfolios are constructed on bivariate dependent sorting using the beta and downside beta. The random forest model was employed as a surrogate for LSTM to detect the nonlinear and threshold-based conditional relationship between risk factors and portfolio returns. The findings of the study reveal that LSTM has robust predictive ability and effectively captures the relationship between risk factors and portfolio returns. The relationship is more conspicuous in high downside risk portfolios than in the medium and high beta portfolio groups. Furthermore, portfolios within the high beta group, regardless of their downside beta levels, exhibit positive returns. The study suggests that investors need to consider high-risk portfolios for stable earnings and predictive returns.

    Citation: Tahir Afzal, Muhammad Asim Afridi, Muhammad Naveed Jan. Integrating LSTM with Fama-French six factor model for predicting portfolio returns: Evidence from Shenzhen stock market China[J]. Data Science in Finance and Economics, 2025, 5(2): 177-204. doi: 10.3934/DSFE.2025009

    Related Papers:

    [1] Yoichi Enatsu, Yukihiko Nakata . Stability and bifurcation analysis of epidemic models with saturated incidence rates: An application to a nonmonotone incidence rate. Mathematical Biosciences and Engineering, 2014, 11(4): 785-805. doi: 10.3934/mbe.2014.11.785
    [2] Jinzhe Suo, Bo Li . Analysis on a diffusive SIS epidemic system with linear source and frequency-dependent incidence function in a heterogeneous environment. Mathematical Biosciences and Engineering, 2020, 17(1): 418-441. doi: 10.3934/mbe.2020023
    [3] Wenzhang Huang, Maoan Han, Kaiyu Liu . Dynamics of an SIS reaction-diffusion epidemic model for disease transmission. Mathematical Biosciences and Engineering, 2010, 7(1): 51-66. doi: 10.3934/mbe.2010.7.51
    [4] Baoxiang Zhang, Yongli Cai, Bingxian Wang, Weiming Wang . Dynamics and asymptotic profiles of steady states of an SIRS epidemic model in spatially heterogenous environment. Mathematical Biosciences and Engineering, 2020, 17(1): 893-909. doi: 10.3934/mbe.2020047
    [5] Yanyan Du, Ting Kang, Qimin Zhang . Asymptotic behavior of a stochastic delayed avian influenza model with saturated incidence rate. Mathematical Biosciences and Engineering, 2020, 17(5): 5341-5368. doi: 10.3934/mbe.2020289
    [6] James M. Hyman, Jia Li . Epidemic models with differential susceptibility and staged progression and their dynamics. Mathematical Biosciences and Engineering, 2009, 6(2): 321-332. doi: 10.3934/mbe.2009.6.321
    [7] Ruixia Zhang, Shuping Li . Analysis of a two-patch SIS model with saturating contact rate and one- directing population dispersal. Mathematical Biosciences and Engineering, 2022, 19(11): 11217-11231. doi: 10.3934/mbe.2022523
    [8] Abdennasser Chekroun, Mohammed Nor Frioui, Toshikazu Kuniya, Tarik Mohammed Touaoula . Global stability of an age-structured epidemic model with general Lyapunov functional. Mathematical Biosciences and Engineering, 2019, 16(3): 1525-1553. doi: 10.3934/mbe.2019073
    [9] Fang Wang, Juping Zhang, Maoxing Liu . Dynamical analysis of a network-based SIR model with saturated incidence rate and nonlinear recovery rate: an edge-compartmental approach. Mathematical Biosciences and Engineering, 2024, 21(4): 5430-5445. doi: 10.3934/mbe.2024239
    [10] Peter Rashkov, Ezio Venturino, Maira Aguiar, Nico Stollenwerk, Bob W. Kooi . On the role of vector modeling in a minimalistic epidemic model. Mathematical Biosciences and Engineering, 2019, 16(5): 4314-4338. doi: 10.3934/mbe.2019215
  • The study aimed to examine the effectiveness of long short-term memory (LSTM) model in predicting portfolio returns employing Fama and French's six-factor model. Monthly data were collected for A-share prices for firms listed on the Shenzhen Stock Exchange, China, for the period extended over 25 years (1997–2022). Portfolios are constructed on bivariate dependent sorting using the beta and downside beta. The random forest model was employed as a surrogate for LSTM to detect the nonlinear and threshold-based conditional relationship between risk factors and portfolio returns. The findings of the study reveal that LSTM has robust predictive ability and effectively captures the relationship between risk factors and portfolio returns. The relationship is more conspicuous in high downside risk portfolios than in the medium and high beta portfolio groups. Furthermore, portfolios within the high beta group, regardless of their downside beta levels, exhibit positive returns. The study suggests that investors need to consider high-risk portfolios for stable earnings and predictive returns.



    Energy is one of the needs of human life. Increasing energy demand is an increasing prosperity indicator, but there are still problems in the energy supply business. Energy use increases with population growth. Large energy consumption causes energy availability to run low. Electricity consumption in Indonesia reaches 202,845.82 GWh. It is a quite large nominal compared to some other countries in Southeast Asia.

    Indonesia is a tropical country that has considerable solar energy potential. The potential of solar energy in Indonesia is around 4.8 kWh/m2/day. There are two types of technology used to exploit the potential of solar energy, namely thermal solar energy technology and solar panel energy. Solar panel energy is used to meet the needs of electricity, water pumps, televisions, telecommunications, and refrigerators in remote areas. The alternative electrical energy produced by solar panels is usually stored first in a storage system (battery) before use. Solar panels become a renewable energy source that offers many advantages, including the energy used, comes from the sun, does not produce pollution, low maintenance costs, and does not produce noise.

    The conversion of solar energy into electrical energy depends on the angle of reception of solar radiation. There are two types of use of panels, namely the PV fixed and solar tracker. Nowadays a lot of optimization has been done on solar panels, with solar tracking systems. The benefit of solar tracker systems is to optimize the collection of solar energy [1]. There are two types of solar tracking based on the axis of the system, namely a single axis and double axis [2,3,4]. The efficiency of dual-axis solar tracking systems is known to be 81.68% higher than in fixed panels, while the efficiency of single-axis tracking is only 32.17% higher than in fixed panels [5]. Solar tracking research focuses on adding the Maximum Power Point Tracking (MPPT) algorithm to the PV panel system. The output power of PV is influenced by the level of solar irradiation and the temperature received by the PV panel surface. If the irradiation received by the PV panel is higher then the resulting current will be higher, but if the panel temperature is too high the voltage obtained will be lower. There have been many studies on MPPT, one of which uses fuzzy Particle Swarm Optimization (PSO) [6]. From the results of experiments and simulations it is known that tracking time and accuracy increase by 0.88% and 0.93% of fuzzy without PSO [7].

    The construction of solar power plants requires very good planning so as not to cause excessive impact on the environment. To build a solar power plant requires an initial investment that is quite expensive, so it needs to be designed components of solar power plants, one of them is a solar charge controller that is reliable, optimal, efficient, and economical so that solar power plant development planning obtained efficiently and optimally. The sepic converter already design in 2008 without a controller [8]. The MPPT algorithm using Perturb and Observe (P & O) technique can get maximum power from the PV panel so that it can charge the battery to the maximum [9,10]. The battery charging using a fuzzy logic controller can track setpoint with a minimum time [11]. The method can resolve problem P & O technique with better result or more accurate. The MPPT based on Particle Swarm Fuzzy can increase the power output of solar tracker than using MPPT based fuzzy [6]. In this research, apply the MPPT Particle Swarm Fuzzy on the battery charging system of solar tracker. The purpose of this research can absorb maximum energy from the sun to the charging system. The design of a charging system controller where the setpoint charge will adjust to the state of the battery voltage so that this controller will work efficiently and optimally. This research is expected to be able to good contribute to solar power plants development planning.

    In this research, the PV module used is a type of polycrystalline solar module with a nominal power of 250 Wp. Table 1 presents the PV specifications as provided by the manufacturer datasheet.

    Table 1.  Solar panels specifications.
    Model Polycrystalline
    Maximum Power at STC (PMPP) 250 W
    Maximum Power Voltage at STC (VMPP) 30 V
    Maximum Power Current at STC (IMPP) 8.34 A
    Open Circuit Voltage at STC (VOC) 36.8 V
    Short Circuit Current at STC (ISC) 9 A
    Temperature Coefficient of VOC -0.32%/℃
    Temperature Coefficient of ISC 0.06%/℃
    Series Connected Cell per Modul 60

     | Show Table
    DownLoad: CSV

    Figure 1 is a characteristic curve of solar panels resulting from simulation under Standard Temperature Condition (STC).

    Figure 1.  Characteristic curve of solar panels.

    Meanwhile, the converter used is a Single-Ended Primary-Inductor (SEPIC) type converter. The Sepic converter is capable of producing above or below the input voltage without reversing the input voltage pole.

    Figure 2 explains a SEPIC converter circuit with the component parameters. The component parameter value of a SEPIC converter can show in Table 2.

    Figure 2.  SEPIC circuit.
    Table 2.  Component parameters of the SEPIC converter.
    Component Simulation Value Experiment Value Unit
    C1 220 220 (50) uF (V)
    L1 800 760 uH
    C3 10 10 (450) uF (V)
    L2 800 760 uH
    D 5 40 A
    C2 3300 3300 (50) uF (V)
    MOSFET IRF540 IRF540 -
    Freq. 50000 50000 Hz

     | Show Table
    DownLoad: CSV

    After entering the SEPIC converter parameter values in the circuit, the circuit modeled on PSIM as shown Figure 3. Converter modeling in PSIM aims to determine the voltage and current characteristics through the component without taking into account the specifications used. Figure 4 explains the current characteristics of SEPIC circuit.

    Figure 3.  SEPIC Modeling on PSIM.
    Figure 4.  Current characteristics through the C1, L1, C2, L2, and R on SEPIC circuit.

    Figure 5 shows the results of the SEPIC converter modeling response. The result of the input voltage is 30 V, while the output voltage is 29 V. The SEPIC converter can produce lower or higher output voltages than the input voltage. The output voltage polarity is directly proportional to the input voltage. It shows that the modeling is already appropriate with existing conditions. The MPPT Algorithm using the P & O technique can only get Maximum Power Point (MPP) on the converter. The MPPT based fuzzy logic controller can maintain the MPP even though there is the load changes. Increased power generated by MPPT sought by optimizing fuzzy membership using PSO. This method can increase maximum power than using fuzzy MPPT [6].

    Figure 5.  Characteristics of input voltage (Red) and output voltage (Blue).

    Primary data collection includes data irradiation, panel temperature, voltage, and current. The 12 Ah battery function to load that first connected to the Solar Charge Controller (SCC). The data collected on 17 April 2018 at 3rd floor of the S2 building in the Department of Physics Engineering ITS start from 6:00 to 17:00 WIB. This location avoid from shading that can obstruct irradiation to the solar panel surface. Irradiation measurements are carried out using a pyranometer. Panel temperature measurements are performed using an infrared thermometer. Current and voltage measurements are carried out with a digital multimeter. The retrieval data results of solar irradiation and panel temperatures for one day (11 hours of operation) can show in Table 3. The highest panel irradiation and temperature obtain at 11:30 with a value of 1011,333 W/m2 and 61,667 ℃. Meanwhile, the maximum power of solar panel occurs at 12:30 with value 196.99 watt or 28.86 volt and 6.826 ampere. The measurement result can be influenced by some factor as wind, wheater, solar panel position, etc at the time.

    Table 3.  Data of Irradiation, Temperature, Current and Voltage on Solar Panel.
    Time Irradiation (W/m2) Temperature (℃) Voltage (Volt) Current (Ampere)
    6:00 31.5 26.867 28.62 0.251
    6:30 59.367g 27.667 29.733 0.53
    7:00 194.9 31.8 30.86 1.524
    7:30 343.7 50.667 30.667 3.093
    8:00 419.5 41.2 30.6 3.364
    8:30 263.067 41.533 29.43 2.2
    9:00 709.933 47.6 30.55 5.43
    9:30 227.533 41.6 29.92 1.93
    10:00 923.933 55 30.25 6.267
    10:30 946.4 57.267 29.576 5.667
    11:00 1000.967 59.133 29.66 6.633
    11:30 1011.333 61.667 29.25 5.76
    12:00 990.333 58.267 29.333 5.813
    12:30 938.933 59.533 28.86 6.826
    13:00 912.167 55.867 29.336 6.75
    13:30 793.767 57.533 29.133 6.6
    14:00 706.933 58.2 29.62 4.446
    14:30 518.8 48.2 30.16 3.822
    15:00 74.1 36.4 28.65 0.619
    15:30 67.467 33.533 28.8 0.562
    16:00 56.333 33.8 28.41 0.445
    16:30 33.867 32.667 27.51 0.266
    17:00 13 30.4 25.76 0.1

     | Show Table
    DownLoad: CSV

    There are three elements of fuzzy control, namely fuzzification, inference system, and defuzzification. The fuzzy control system is used to track the output of the SEPIC converter to match the battery's charging voltage. Inputs given to the controller are voltage errors (error) and voltage delta error (derror), while the output is the duty cycle (D). The voltage error is the difference between the setpoint and the actual voltage, while voltage delta error is the difference between the actual voltage and the previous voltage. The duty cycle of Pulse Width Modulation (PWM) is the pulse width used to set the switching speed on the converter. To set the duty cycle, a Linear Decreasing Weight Particle Swarm Optimization (LDWPSO) control is needed [12]. The Membership Function (MF) used is the type of triangle and trapezoid with 3 MF triangles and 2 MF trapezoid. The MPSO (Modified Particle Swarm Optimization) used is LDWPSO. LDWPSO is a modification of PSO which modifies the weight of inertia to decrease with each additional iteration. Figure 6 is the MF of fuzzy input and output that has been optimized by LDWPSO.

    Figure 6.  Membership function of Fuzzy-LDWPSO.

    The MPPT with the P & O technique only has two inputs. If the voltage is less than zero, then the PWM will go up. If the voltage is more than zero, then the PWM will go down. Meanwhile, the fuzzy logic controller has every two inputs and one output divided by the five membership functions of fuzzy. This method can increase the power at the PV output rather than the P & O technique. The MPPT fuzzy MPSO is applied to control the output of the converter [6]. The rule base is used to control the pulse width with two inputs that are errors and delta errors. The rule base design of fuzzy control can explain in Table 4. The rule base that has been designed using the AND operator. For example, a rule (1), if the error value enters the Negative Big (NB) membership function range and the delta error enter Negative Big (NB) membership function range, then the duty cycle value enter Negative Big (NB) membership function range. If occur errors and delta errors are very big, then the duty cycle of PWM will go up very fast. (IF e(NB) AND de(NB) THEN duty cycle(PB) etc and applies to all rules).

    Table 4.  Rule base fuzzy.
    NB N Z P PB
    NB PB (1) PB (2) P (3) Z (4) Z (5)
    N P (6) P (7) Z (8) Z (9) Z (10)
    Z Z (11) Z (12) Z (13) Z (14) Z (15)
    P Z (16) Z (17) N (18) N (19) N (20)
    PB N (21) N (22) N (23) NB (24) NB (25)

     | Show Table
    DownLoad: CSV

    Characteristics test of the SEPIC converter designed with several variations of the switching frequency and duty cycle. The purpose of this test is to determine the characteristics of the converter. Frequency variation testing is performed to determine the effect of switching frequencies used on the output voltage. Frequency Switching testing is carried out in the range of 10,000 Hz–60,000 Hz. The tests use a DC voltage source in the form of a power generator (16 V/2 A), a 6.8 Ohm converter load and a 30% duty cycle.

    Table 5 shows the effect of switching frequency on the SEPIC converter output. NA (Not Available) is a condition where the generator cannot produce the desired input voltage. The switching frequency not visible significantly influence the output voltage produced by the converter. The converter starts properly working when given a switching frequency of 30 kHz or more.

    Table 5.  Effect of switching frequency on the converter output voltage.
    Vin Frequency Switching
    10 k 20 k 30 k 40 k 50 k 60 k
    1 0.15 0.11 0.13 0.13 0.13 1.4
    2 0.45 0.49 0.48 0.5 0.48 0.49
    3 0.79 0.86 0.83 0.88 0.84 0.86
    4 1.16 1.28 1.2 1.26 1.23 1.21
    5 1.56 1.6 1.6 1.64 1.64 1.6
    6 2.5 2.03 1.96 2.04 2.02 1.96
    7 2.87 2.37 2.35 2.38 2.42 2.32
    8 NA 2.73 2.74 2.8 2.82 2.81
    9 NA 3.11 3.17 3.17 3.21 3.2
    10 NA 3.52 3.53 3.56 3.62 3.6
    11 NA 3.94 3.89 3.94 4.01 3.95
    12 NA 4.5 4.35 4.33 4.43 4.45
    13 NA NA 4.75 4.7 4.84 4.8
    14 NA NA 5.18 5.1 5.23 5.25
    15 NA NA 5.6 5.53 5.61 5.64
    16 NA NA 6 6.02 6.04 6

     | Show Table
    DownLoad: CSV

    The duty cycle variation aims to determine the duty cycle effect on the output voltage generated by the converter. The tests use a DC generator (16 V/2 A), a resistor load of 6.8 Ohm and a switching frequency of 50 kHz. The results of testing the duty cycle variations shown in Table 6. It shows the SEPIC converter output voltage response with duty cycle variations. The output voltage value is high if the duty cycle value is also high.

    Table 6.  Effect of duty cycle on the converter output voltage.
    Vin Duty Cycle
    10% 20% 30% 40% 50%
    1 0.01 0.03 0.12 0.31 0.32
    2 0.03 0.2 0.49 0.87 0.92
    3 0.09 0.41 0.83 1.46 1.52
    4 0.16 0.62 1.17 2.03 2.09
    5 0.26 0.85 1.56 2.61 2.64
    6 0.35 1.11 1.93 3.21 3.22
    7 0.45 1.38 2.3 3.73 3.81
    8 0.55 1.56 2.67 4.25 4.37
    9 0.64 1.8 3.04 4.89 5.02
    10 0.74 2 3.45 5.42 5.63
    11 0.84 2.24 3.8 6.05 6.21
    12 0.94 2.5 4.13 6.65 6.75

     | Show Table
    DownLoad: CSV

    The function of tracking simulation is to find out the comparison between fuzzy controller and fuzzy-LDWPSO that has a design for tracking setpoint charging (14.55 V). The simulation is carried out on STC conditions, namely irradiation conditions of 1000 W/m2 and Temperature 25 ℃. Figure 7 shows the red line is SEPIC converter output voltage using fuzzy control. The blue line is the output voltage with fuzzy-LDWPSO control while the green line is setpoint voltage 14.55 V. The voltage ratio experienced by the fuzzy control is in the numbers 15.8 V to 19.8 V while the fuzzy-LDWPSO has an output voltage principle at 14.7 V to 16.3 V.

    Figure 7.  Tracking voltage 14.55 Volts under STC conditions, Fuzzy Control (Red) FLDWPSO Control (Blue).

    Table 7 shows that fuzzy-LDWPSO control has a faster response than fuzzy control. It can be seen based on the decrease in rise time, settling time, and peak time. The maximum overshoot and steady-state error values of the fuzzy-LDWPSO control are smaller than fuzzy. Based on the data in Table 7, fuzzy-LDWPSO is better in tracking the charging voltage by reducing the maximum overshoot by 16.14% and reducing the steady-state error by 27.49% from the fuzzy control.

    Table 7.  Comparison parameters between fuzzy and Fuzzy-LDWPSO control.
    Parameter Fuzzy Fuzzy-LDWPSO
    Rise Time 0.0027 seconds 0.0021
    Settling Time 0.0189 seconds 0.0135
    Maximum Overshoot 43.64% (20.9 V) 27.5% (18.7 V)
    Peak Time 0.008 seconds 0.0056
    Error Steady state (0.11 seconds) 36.08% (19.8 V) 8.59% (15.8 V)

     | Show Table
    DownLoad: CSV

    The test is performed to compare between fuzzy and fuzzy-LDWPSO control in tracking setpoint voltage with variations conditions of temperature and irradiation. Temperature and irradiation variations are used based on the results of secondary irradiation and temperature data collection on 17 April 2018. Figure 8 shows the results of irradiation measurements for 6 hours, start from 6.00 WIB until 17.00 WIB. Based on the graph, the biggest irradiation occurred at 11:30 WIB. The panel surface temperature measured for 10.5 hours shown in Figure 9. The highest panel temperature reached at 11:30 with the temperature recorded on the measurement of 61.66 ℃. The charging voltage tracking test is performed to determine the performance of the fuzzy and fuzzy-LDW-PSO controls that designed. The results of the tracking setpoint test with variations in climatic conditions shown in Figure 10.

    Figure 8.  Irradiation data variations.
    Figure 9.  Temperature data variations.
    Figure 10.  Fuzzy tracking test results (red) and Fuzzy-LDWPSO (blue) with climatic conditions variation.

    In Figure 10, the red line is the results of the output voltage control generated by fuzzy control while the blue line is the fuzzy-LDWPSO output voltage. The voltage ripple that generated Fuzzy located between 14.8 V and 15 V, while the ripple voltage of fuzzy-LDWPSO is in the numbers 14.6 V to 14.8 V. The charging voltage using the fuzzy-LDWPSO control is closer to the setpoint than the fuzzy control. At 15.00, the voltage decreases below the set point because when the irradiation data is collected and the temperature at that hour the sun is covered with thin clouds, so the control cannot track the setpoint voltage.

    Hardware testing for fix and tracking conditions was carried out on 1 June 2018 in the car park area of the ITS Physics Engineering Department. The tests use 2 PVs with a capacity of 250 WP each with fix and track conditions. The function of retrieval data using SCC is to compare with the converter that has designed. Data collected during sunny cloudy.

    Figure 11.  Comparison of charging currents with Fuzzy LDWPSO SEPIC Fix, SEPIC Tracker, SCC Fix, and SCC Tracker.

    The charging current that enters using SCC is very volatile, so it records the maximum and minimum currents that are read by the ampere meter. The charging current entering using the SEPIC experiences a ripple of around ± 0.01 A that read on the ampere meter. PV with a tracking condition produces a charging current is greater than the PV condition fix. The increase in charging current generated by the tracking conditions using the SEPIC converter with fuzzy-LDWPSO Control was 36.15% and using SCC increased by 11.61% from the fix condition panel.

    To find out the response of the fuzzy output can be approached by finding the value of rise time, settling time, maximum overshoot, peak time, and steady-state error. After knowing the results of their responses, we can find out which controller has a good response with the smallest error. The fuzzy control has a value of rise time 0.0027 seconds, settling time 0.0189 seconds, maximum overshoot 43.64%, peak time 0.008 and steady-state error 36.08%. fuzzy-LDWPSO control has a value of rise time 0.0021 seconds, settling time 0.0135 seconds, maximum overshoot 27.5%, peak time 0.0056 seconds, and steady-state error of 8.59%. The fuzzy-LDWPSO control has a better [13] response than the fuzzy controller with response faster and has error smaller. After getting the best controller, the fuzzy-LDWPSO control entered to battery charging system. The SEPIC charging circuit uses a fuzzy-LDWPSO control with a panel tracking condition capable of increasing the battery charging current by 36.15% from the panel condition fix.

    All authors declare no conflicts of interest in this paper.



    [1] Agarwal PK, Pradhan HK (2018) Mutual fund performance using unconditional multifactor models: Evidence from India. J Emerg Mark Financ 17: 157–184. https://doi.org/10.1177/0972652718777056 doi: 10.1177/0972652718777056
    [2] Aharoni G, Grundy B, Zeng Q (2013). Stock returns and the Miller Modigliani valuation formula: Revisiting the Fama French analysis. J Financ Econ 110: 347–357. https://doi.org/10.1016/j.jfineco.2013.08.003 doi: 10.1016/j.jfineco.2013.08.003
    [3] Aminimehr A, Raoofi A, Aminimehr A, et al. (2022) A comprehensive study of market prediction from efficient market hypothesis up to late intelligent market prediction approaches. Comput Econ 60: 781–815. https://doi.org/10.1007/s10614-022-10283-1 doi: 10.1007/s10614-022-10283-1
    [4] Ang A, Chen JS, Xing Y (2002) Downside correlation and expected stock returns. EFA 2002 Berlin Meetings. https://doi.org/10.2139/ssrn.282986
    [5] Ashwood AJ (2013) Portfolio selection using artificial intelligence, Queensland University of Technology. Available from: https://eprints.qut.edu.au/66229/1/Andrew_Ashwood_Thesis.pdf.
    [6] Ayub U, Kausar S, Noreen U, et al. (2020) Downside risk-based six-factor capital asset pricing model (CAPM): A new paradigm in asset pricing. Sustainability 12: 6756. https://doi.org/10.3390/SU12176756 doi: 10.3390/SU12176756
    [7] Bhandari HN, Rimal B, Pokhrel NR, et al. (2022) Predicting stock market index using LSTM. Mach Learn Appl 9: 100320. https://doi.org/10.1016/j.mlwa.2022.100320 doi: 10.1016/j.mlwa.2022.100320
    [8] Buditomo B, Candra S, Soetanto TV (2024) Fama and french five-factor study of stock market in indonesia. Int J Organ Behav Policy 3: 39–52. https://doi.org/10.9744/ijobp.3.1.39-52 doi: 10.9744/ijobp.3.1.39-52
    [9] Carhart MM (1997) On persistence in mutual fund performance. J Financ 52: 57–82. https://doi.org/10.1111/j.1540-6261.1997.tb03808.x doi: 10.1111/j.1540-6261.1997.tb03808.x
    [10] Carpenter JN, Whitelaw RF (2017) The development of China's stock market and stakes for the global economy. Ann Rev Financ Econ 9: 233–257. https://doi.org/10.1146/annurev-financial-110716-032333 doi: 10.1146/annurev-financial-110716-032333
    [11] Cederburg S, O'Doherty MS (2015) Asset-pricing anomalies at the firm level. J Econ 186: 113–128. https://doi.org/https://doi.org/10.1016/j.jeconom.2014.06.004 doi: 10.1016/j.jeconom.2014.06.004
    [12] Daniel K, Moskowitz TJ (2014) Momentum crashes. NATIONAL BUREAU OF ECONOMIC RESEARCH. Available from: http://www.nber.org/papers/w20439.
    [13] Diallo B, Bagudu A, Zhang Q (2023) Fama–French three versus five, which model is better? A machine learning approach. J Forecast 42: 1461–1475. https://doi.org/https://doi.org/10.1002/for.2970 doi: 10.1002/for.2970
    [14] Dirkx P, Peter FJ (2020) The Fama-French five-factor model plus momentum: Evidence for the german market. Schmalenbach Bus Rev 72: 661–684. https://doi.org/10.1007/s41464-020-00105-y doi: 10.1007/s41464-020-00105-y
    [15] Estrada J (2007) Mean-semivariance behavior: Downside risk and capital asset pricing. Int Rev Econ Financ 16: 169–185. https://doi.org/10.1016/j.iref.2005.03.003 doi: 10.1016/j.iref.2005.03.003
    [16] Estrada J, Serra AP (2005) Risk and return in emerging markets: Family matters. J Multinational Financ Manag 15: 257–272. https://doi.org/10.1016/j.mulfin.2004.09.002 doi: 10.1016/j.mulfin.2004.09.002
    [17] Fama EF, French KR (1992) The cross‐section of expected stock returns. J Financ 47: 427–465. https://doi.org/10.1111/j.1540-6261.1992.tb04398.x doi: 10.1111/j.1540-6261.1992.tb04398.x
    [18] Fama EF, French KR (2006) Profitability, investment and average returns. J Financ Econ 82: 491–518. https://doi.org/10.1016/j.jfineco.2005.09.009 doi: 10.1016/j.jfineco.2005.09.009
    [19] Fama EF, French KR (2015) A five-factor asset pricing model. J Financ Econ 116: 1–22. https://doi.org/10.1016/j.jfineco.2014.10.010 doi: 10.1016/j.jfineco.2014.10.010
    [20] Fama EF, French KR (2018) Choosing factors. J Financ Econ 128: 234–252. https://doi.org/10.1016/j.jfineco.2018.02.012 doi: 10.1016/j.jfineco.2018.02.012
    [21] Feng G, He J, Polson NG (2018) Deep learning for predicting asset returns. arXiv Preprint 1804: 1–23.
    [22] Feng Z, Runtong Z, Lingyun H (2007) A prospective study of the price behaviors of Chinese stock markets. Chinese Control Conference, 558–562.
    [23] Galagedera DU (2009) An analytical framework for explaining relative performance of CAPM beta and downside beta. Int J Theor Appl Financ 12: 341–358. https://doi.org/10.1142/S0219024909005257 doi: 10.1142/S0219024909005257
    [24] Giglio S, Kelly B, Xiu D (2022) Factor models, machine learning, and asset pricing. Ann Rev Financ Econ 14: 337–368. https://doi.org/10.1146/annurev-financial-101521-104735 doi: 10.1146/annurev-financial-101521-104735
    [25] Goo Y, Wang C (2024) Do there exist nonlinear phenomena of the Fama-French six factors on stock returns? An empirical investigation on the Taiwan stock market. Modern Econ 15: 394–412. https://doi.org/10.4236/me.2024.154021 doi: 10.4236/me.2024.154021
    [26] Hall P, Gill N (2019) An introduction to machine learning interpretability. O'Reilly Media, Incorporated.
    [27] Hochreiter S, Schmidhuber J (1997) Long short-term memory. Neural Compu 9: 1735–1780.
    [28] Hu GX, Chen C, Shao Y, et al. (2019) Fama–French in China: Size and Value factors in Chinese stock returns. Int Rev Financ 19: 3–44. https://doi.org/10.1111/irfi.12177 doi: 10.1111/irfi.12177
    [29] Hu GX, Pan J, Wang J (2018) Chinese capital market: An empirical overview. NBER WORKING PAPER SERIES. Available from: https://www.nber.org/system/files/working_papers/w24346/w24346.pdf.
    [30] Jegadeesh N, Titman S (1993) Returns to buying winners and selling losers: Implications for stock market efficiency. J Financ 48: 65–91. https://doi.org/10.2307/2328882 doi: 10.2307/2328882
    [31] Jia H (2023) Comparison on Asset Pricing Models: Application of CAPM, Fama-French 3 Factors model and Fama-French 5 factors model. BCP Bus Manag 40: 119–127. https://doi.org/10.54691/bcpbm.v40i.4367 doi: 10.54691/bcpbm.v40i.4367
    [32] Jinghan C (2010) Trading preference of Chinese investors and its impact on stock volatility. Rev Financ Stud 3: 53–64.
    [33] Joshi V (2024) Cross-section of expected stock returns: An application of Fama-French Five Factor Model in Nepal. Batuk 10: 72–89. https://doi.org/10.3126/batuk.v10i1.62299 doi: 10.3126/batuk.v10i1.62299
    [34] Kazmi M, Noreen U, Jadoon IA, et al. (2021) Downside beta and downside gamma: In search for a better capital asset pricing model. Risks 9: 223. https://doi.org/10.3390/risks9120223 doi: 10.3390/risks9120223
    [35] Kundlia S, Verma D (2021) Illiquidity premium in the Indian stock market: An empirical study. Asian Econ Financ Rev 11: 501–511. https://doi.org/10.18488/JOURNAL.AEFR.2021.116.501.511 doi: 10.18488/JOURNAL.AEFR.2021.116.501.511
    [36] Lintner J (1965) Security prices, risk, and maximal gains from diversification. J Financ 20: 587–615. https://doi.org/10.2307/2977249 doi: 10.2307/2977249
    [37] Liu J (2019) A novel downside risk measure and expected returns. The World Bank. https://doi.org/10.2139/ssrn.3406944 doi: 10.2139/ssrn.3406944
    [38] Liu X, Wang X (2023) Stock investment strategies with time-varying co-movements between energy market and global financial market. Highlights Busi Econ Manag 21: 862–866.
    [39] Markowitz H (1952) Portfolio selection. J Financ 7: 77–91.
    [40] Mahpudin E (2020) The Effect of Macroeconomics on Stock Price Index in the Republic of China. Int J Econ Bus Adm VIII: 228–236. https://doi.org/10.35808/ijeba/511
    [41] Miller MH, Modigliani F (1961) Dividend policy, growth, and the valuation of shares. J Bus 34: 411. https://doi.org/10.1086/294442 doi: 10.1086/294442
    [42] Miralles-Quirós JL, Miralles-Quirós MM, Nogueira JM (2020) Sustainable development goals and investment strategies: The profitability of using Five-Factor Fama-French alphas. Sustainability 12: 1–16. https://doi.org/10.3390/su12051842 doi: 10.3390/su12051842
    [43] Molnar C (2020) Interpretable machine learning. Lulu. com.
    [44] Mossin J (1966) Equilibrium in a capital asset market. Econometrica 34: 768–783. https://doi.org/10.2307/1910098 doi: 10.2307/1910098
    [45] Novy-Marx R (2013) The other side of value: The gross profitability premium. J Financ Econ 108: 1–28. https://doi.org/10.1016/j.jfineco.2013.01.003 doi: 10.1016/j.jfineco.2013.01.003
    [46] Petrozziello A, Troiano L, Serra A, et al. (2022) Deep learning for volatility forecasting in asset management. Soft Comput 26: 8553–8574. https://doi.org/10.1007/s00500-022-07161-1 doi: 10.1007/s00500-022-07161-1
    [47] Sharma K, Bhalla R (2022) Stock market prediction techniques: A review paper. Second International Conference on Sustainable Technologies for Computational Intelligence, 175–188. https://doi.org/10.1007/978-981-16-4641-6_15 doi: 10.1007/978-981-16-4641-6_15
    [48] Sharpe WF (1964) Capital asset prices: A theory of market equilibrium under conditions of risk. J Financ 19: 425–442. https://doi.org/10.1111/j.1540-6261.1964.tb02865.x doi: 10.1111/j.1540-6261.1964.tb02865.x
    [49] Sharpe WF (1966) Mutual fund performance. J Bus 39: 119–138. http://www.jstor.org/stable/2351741
    [50] Singh K, Singh A, Prakash P (2023) Testing factor models in an emerging market: Evidence from India. Int J Manag Financ 19: 203–232. https://doi.org/10.1108/IJMF-05-2021-0245 doi: 10.1108/IJMF-05-2021-0245
    [51] Sortino FA, Van Der Meer R (1991) Downside risk. J Portfolio Manag 17: 27.
    [52] Staudemeyer RC, Morris ER (2019) Understanding LSTM -- a tutorial into long short-term memory Recurrent Neural Networks. arXiv Preprint 1909: 09586. http://arXiv.org/abs/1909.09586
    [53] Sun J, Manger G (2012) Listing on the Shenzhen Stock Exchange: Behavioural finance implications. ASBBS Proc 19: 874–889. http://search.proquest.com/docview/1445142469?accountid = 79789
    [54] Tan L, Zhang X, Zhang X (2024) Retail and institutional investor trading behaviours: Evidence from China. Ann Rev Financ Econ 16: 459–483.
    [55] Yang P, Yang L (2022) Asset pricing and nominal price illusion in China. Humanit Soc Sci Communicat 9: 1–9. https://doi.org/10.1057/s41599-022-01133-4 doi: 10.1057/s41599-022-01133-4
    [56] Zaghouani CA, Mezzez HD (2020). Evidence on aggregate volatility risk premium for the French Stock Market. Managerial Financ 46: 72–91. https://doi.org/10.1108/MF-11-2018-0535 doi: 10.1108/MF-11-2018-0535
    [57] Zhang C (2022) Asset pricing and deep learning. arXiv Preprint 2209: 1–18. http://arXiv.org/abs/2209.12014
    [58] Zheng Y, Wen F, Deng H, et al. (2022) The relationship between carbon market attention and the EU CET market: Evidence from different market conditions. Financ Res Lett 50: 103140. https://doi.org/https://doi.org/10.1016/j.frl.2022.103140 doi: 10.1016/j.frl.2022.103140
  • DSFE-05-02-009-S001.pdf
  • This article has been cited by:

    1. Giovanni Susinni, Santi Agatino Rizzo, Francesco Iannuzzo, Two Decades of Condition Monitoring Methods for Power Devices, 2021, 10, 2079-9292, 683, 10.3390/electronics10060683
    2. Zhang Yan, Zhang Miyuan, Wang Yajun, Cai Xibiao, Li Yanjun, Photovoltaic MPPT algorithm based on adaptive particle swarm optimization neural-fuzzy control, 2023, 44, 10641246, 341, 10.3233/JIFS-213387
    3. Hang Cheng, Shaowu Li, Zhipeng Fan, Liwei Liu, 2021, Intelligent MPPT Control Methods for Photovoltaic System: A review, 978-1-6654-4089-9, 1439, 10.1109/CCDC52312.2021.9602802
    4. Imam Abadi, Ali Musyafa, Kurnia Diastana Abdirandra, Dwi Nur Fitriyanah, 2023, 2808, 0094-243X, 040001, 10.1063/5.0126518
    5. R R Manullang, M Yanti, C D Hasibuan, E Yudhistira, Performance of Secure Door Access System Using Identification Numbers, Fingerprints, and Facial Recognition, 2024, 2733, 1742-6588, 012003, 10.1088/1742-6596/2733/1/012003
    6. Asrori Asrori, Yuniarto Winoko, Subagiyo Subagiyo, Pondi Udianto, Irwan Eryk, Design and development of hybrid solar e-bike for sustainable green transportation, 2023, 21, 1451-4117, 1139, 10.5937/jaes0-45297
  • Reader Comments
  • © 2025 the Author(s), licensee AIMS Press. This is an open access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0)
通讯作者: 陈斌, bchen63@163.com
  • 1. 

    沈阳化工大学材料科学与工程学院 沈阳 110142

  1. 本站搜索
  2. 百度学术搜索
  3. 万方数据库搜索
  4. CNKI搜索

Metrics

Article views(441) PDF downloads(25) Cited by(0)

Figures and Tables

Figures(7)  /  Tables(4)

/

DownLoad:  Full-Size Img  PowerPoint
Return
Return

Catalog