Forecasting the Vietnamese pharmacy company’s stock prices using Statistic, Machine Learning and Deep Learning Models.
This paper investigates the application of various statistical models and machine learning algorithms in forecasting the stock prices of Vietnamese pharmaceutical companies. Leveraging a diverse array of methodologies, including Linear Regression (LR), Support Vector Regression (SVR), Long Short-Term Memory (LSTM) networks, Autoregressive Integrated Moving Average (ARIMA), Autoregressive Integrated Moving Average with Exogenous Variables (ARIMAX), K-Nearest Neighbors (KNN), and Boosting LSTM, our study aims to provide comprehensive insights into the predictive capabilities of these models within the dynamic and complex pharmaceutical stock market landscape. Additionally, the findings contribute to the ongoing discourse on applying statistical models and machine learning algorithms in predicting stock prices within emerging markets, particularly in the context of 3 Vietnamese Pharmaceutical Joint Stock Company (OPC, Vimedimex, and IMEXPHARM).
Keyword: Stock price, forecasting, LR, SVR, LSTM, ARIMA, ARIMAX, KNN, Boosting LSTM, HMM,pharma companies, OPC, IMP, VMD.
We have decided to approach this problems by introducing to you 8 algorithms below includes Statistical Models, Machine Learning and Deep Learning Models.
A multiple linear regression model has the form Eq.(19):
-
AutoRegressive - AR(p) is a regression model with lagged I, until p-th time in the past, as predictors. Eq.\eqref{eq}
$$AR(p)={\ \alpha\ +\ \beta_1y_{t-1}\ +\ \beta_2y_{t-2}+\ \cdots\ +\ \beta_py_{t-p}+\ \varepsilon}$$ -
Integrated I(d) - The difference is taken d times until the original series becomes stationary. Eq.\eqref{eq}
$${I\left(d\right)=\ \Delta y_{t}= y_{t}-y_{t-1}.}$$ -
Moving average MA(q) - A moving average model uses a regression-like model on past forecast errors.\eqref{eq}
$$MA(q)={\ \alpha\ +\ \theta_1\varepsilon_{t-1}\ +\ \theta_2\varepsilon_{t-2}\ +\ \cdots\ +\ \theta_p\varepsilon_{t-p}\ +\ \varepsilon.}$$ -
Regression equations for ARIMA(p,d,q): Eq.\eqref{eq}
$${ARIMA\left(p,q,d\right)=\ AR\left(p\right)+\ I\left(d\right)+\ MA\left(q\right)\ \ \ }$$
K-Nearest Neighbors (KNN) is one of the simpler prediction/classification techniques: there is no model to be fit (as in regression).
H. HIDDEN MARKOVMODEL(HMM)
The model’s observation is a three dimensional vector representing daily stock information,Eq.(19)
The data-set is crawled by using vnstock · PyPI (v0.2.8.6) from Pypi library through DNSC stock market in 10th Dec, 2023.
There are 3 data-set fetching 3 pharma company in Vietnam, includes OPC Pharmaceutical Joint Stock Company (OPC), 2468 rows; VimedimexPharmaceuticalJoint Stock Company (VMD) 2477 rows and IMEXPHARM Pharmaceutical Joint Stock Company (IMP) with 2467 rows from 2nd Jan, 2014 to 8th Dec, 2023.
| Companies | OPC | VMD | IMP |
|---|---|---|---|
| Count | 2,468 | 2,477 | 2,467 |
| Mean | 16,553 | 18,719 | 38,546 |
| Standard deviation | 5,180 | 7,165 | 16,621 |
| Min | 7,470 | 6,140 | 10,790 |
| 25% | 11,172 | 14,950 | 23,280 |
| 50% | 16,700 | 18,150 | 34,890 |
| 75% | 21,760 | 21,000 | 52,075 |
| Max | 27,490 | 77,810 | 79,290 |
| Variance | 26,841,383 | 51,345,375 | 276,290,079 |
| Skewness | -0.14240 | 2.37885 | 0.50166 |
| Kurtosis | -1.35002 | 11.85794 | -0.86655 |
Model Ratio RMSE MAPE MAE
6 -- 3 -- 1 1651.359 6.179 1618.750
LR 7 -- 2 -- 1 1700.792 5.453 1254.153
8 -- 1 -- 1 1603.972 6.365 2218.027
6 -- 3 -- 1 5903.303 25.004 7342.278
ARIMA 7 -- 2 -- 1 1623.326 6.185 1026.099
8 -- 1 -- 1 1175.016 3.855 702.727
6 -- 3 -- 1 4086.551 13.627 3081.360
SVR 7 -- 2 -- 1 2695.950 6.981 1650.327
8 -- 1 -- 1 593.209 0.790 191.922
6 -- 3 -- 1 571.944 1.926 486.944
LSTM 7 -- 2 -- 1 403.939 1.154 197.606
8 -- 1 -- 1 410.994 1.248 209.331
**6 -- 3 -- 1** **200.118** **0.641** **142.276**
ARIMAX 7 -- 2 -- 1 243.033 0.738 165.235
8 -- 1 -- 1 200.118 0.641 142.276
6 -- 3 -- 1 5691.622 24.026 5281.468
KNN 7 -- 2 -- 1 1461.386 4.252 996.334
8 -- 1 -- 1 1192.508 2.444 586.232
6 -- 3 -- 1 724.186 2.229 554.786
Boosting LSTM 7 -- 2 -- 1 592.866 1.544 339.519
8 -- 1 -- 1 737.7 1.878 359.912
6 -- 3 -- 1 4387.945 17.882 3899.961
HMM 7 -- 2 -- 1 12242.473 54.110 12122.201
8 -- 1 -- 1 3054.832 9.839 2213.876
Model Ratio RMSE MAPE MAE
6 -- 3 -- 1 9649.451 16.289 2418.075
LR 7 -- 2 -- 1 12120.135 21.137 2283.258
8 -- 1 -- 1 5714.464 21.078 5916.466
6 -- 3 -- 1 9937.794 18.275 2548.507
ARIMA 7 -- 2 -- 1 12457.71 22.682 2883.917
8 -- 1 -- 1 13172.876 55.797 15002.287
6 -- 3 -- 1 11438.434 12.914 4997.949
SVR 7 -- 2 -- 1 14422.118 22.395 8324.811
8 -- 1 -- 1 735.555 1.864 460.992
6 -- 3 -- 1 1893.718 3.675 474.926
LSTM 7 -- 2 -- 1 1766.212 3.334 333.847
8 -- 1 -- 1 706.622 2.307 308.219
**6 -- 3 -- 1** **406.139** **0.961** **239.240**
ARIMAX 7 -- 2 -- 1 474.067 1.172 309.271
8 -- 1 -- 1 484.918 1.514 356.366
6 -- 3 -- 1 9837.156 18.329 5614.338
KNN 7 -- 2 -- 1 12302.448 22.017 7666.507
8 -- 1 -- 1 12951.406 54.788 11920.842
6 -- 3 -- 1 3295.73 5.526 894.255
Boosting LSTM 7 -- 2 -- 1 4454.588 6.321 614.9697
8 -- 1 -- 1 767.626 2.476 505.563
6 -- 3 -- 1 9964.968 20.851 6136.134
HMM 7 -- 2 -- 1 12427.107 26.853 8515.060
8 -- 1 -- 1 9884.588 31.466 8311.402
Model Ratio RMSE MAPE MAE
6 -- 3 -- 1 14575.349 19.263 5784.508
LR 7 -- 2 -- 1 18488.891 26.043 6404.986
8 -- 1 -- 1 10409.816 12.933 7164.764
6 -- 3 -- 1 15721.151 20.404 17852.991
ARIMA 7 -- 2 -- 1 10833.288 14.187 5648.963
8 -- 1 -- 1 11875.223 16.744 13740.165
6 -- 3 -- 1 30489.105 40.429 25254.876
SVR 7 -- 2 -- 1 31617.288 44.872 29261.537
8 -- 1 -- 1 2624.939 2.308 1541.152
6 -- 3 -- 1 3007.534 4.116 2573.013
LSTM 7 -- 2 -- 1 1809.724 2.337 1054.136
8 -- 1 -- 1 1472.944 1.93 940.072
**6 -- 3 -- 1** **566.148** **0.723** **386.874**
ARIMAX 7 -- 2 -- 1 637.844 0.713 441.870
8 -- 1 -- 1 745.529 0.794 489.264
6 -- 3 -- 1 22355.075 30.649 18843.429
KNN 7 -- 2 -- 1 14449.104 19.35 12783.089
8 -- 1 -- 1 8747.178 12.626 7198.577
6 -- 3 -- 1 2761.74 3.872 1693.198
Boosting LSTM 7 -- 2 -- 1 2501.154 3.167 1511.618
8 -- 1 -- 1 2377.513 3.279 1337.611
6 -- 3 -- 1 23104.035 33.584 20096.218
HMM 7 -- 2 -- 1 26204.521 40.751 25651.172
8 -- 1 -- 1 14171.327 16.793 10559.083
As you can see in table 3,4 and 5 ARIMAX has most less difference between actual value andpredict (RMSE: 200.118vnd,MAPE: 0.641%,MAE: 142.276vnd) value at ratio(6-3-1) for both 3companies. LSTMModelandBoost ing LSTM model rank 2nd (Ratio:8-1-1,RMSE:767.626vnd, MAPE: 2.307%,MAE: 380.219vnd) and 3rd(Ratio: 8-1 - 1,RMSE: 767.626vnd,MAPE: 2.4%,MAE: 505.563vnd) respectively. While ARIMA has big gap between 2compared values (approximately 13k vnd and percent errormostly 56%).