Forecasting the movement of growth, income & value stocks using ARIMA

By Riya Jain & Priya Chetty on July 16, 2021

Stock markets inherently carry uncertainty, leading to risks. Investors, therefore, use statistical models to predict the movement of a stock before investing, so that their returns can be maximized. The previous article examined the nature of the dataset for growth, income, and value stocks over the period 2000 to 2020 using the ARIMA model. ARIMA helped predict the movement of the average closing price and average return. This article creates the optimal model and predicts the performance of different categories of stocks for the next 10 years i.e. from 1st April 2020 to 31st March 2030. Herein, the selection of the optimal model from the stocks is done based on the value of average returns (AR) and moving average (MA) coefficients, significance level, and value of AIC and BIC. The rules for determining the model at a 5% or 10% level of significance level are stated below.

  1. The value of the majority of AR and MA coefficients at different lags should be significant.
  2. The optimal model should have a minimum value for AIC or BIC.
  3. The majority of significant values of AR and MA coefficient should be less than 1.

A model fulfilling all the above-stated conditions would be selected from the given list of ARIMA models to forecast the performance of growth, income, and value stocks.

The modeling of the time series presents the linkage of the current performance of stocks with the historical data.

The average closing price of growth stocks using ARIMA

Results of all possible models identified in the previous article are shown in Table 1. Herein the cells highlighted in yellow colour depict the coefficients having a p-value of more than 0.05 or 0.10. Based on the 2nd condition, the model (1, 2, 2), (3, 2, 1), and (2, 2, 1) has the least AIC or BIC value. The cells with green colour represent the coefficient value of AR and MA wherein the lag is significant and has a coefficient value of less than 1. However, the identified models (1, 2, 2) and (3, 2, 1) has an absence of significant and less than 1 AR and MA coefficient but model (2, 2, 1) has their AR and MA coefficient value significant as well as less than 1. Thus, the ARIMA model (2, 2, 1) is the optimal model for the average closing price.

Table 1: Results of all possible Average closing price models

Based on the coefficient values, the model can be written as

Average closing price forecasting model for growth stocks
Equation 1: Average closing price forecasting model for growth stocks

Where Y represents the average closing price and shows the error term of the model.

The determined model is used for the prediction of the future performance of the average closing price. Herein, the average closing price shows that over time there has been a reduction in the closing price value of the growth stocks as the value has decreased from 1052.874 in FY 2019 to 781.2488 in FY 2020. With a slight recovery in FY 2021, there would again be a decrease in the performance of growth stocks. This examination of the growth stock’s closing price further is shown in Figure 1.

Financial YearAnnual Average Closing Price
20181074.216
20191052.874
2020781.2488
2021807.8962
2022799.9501
2023826.5974
2024808.1012
Table 2: Annual closing price of growth stocks
Predicted closing price of growth stocks
Figure 1: Predicted closing price of growth stocks

The above figure shows that there would be high variation in the performance of growth stocks but the upward trend is seen in the year 2024 wherein investors would have an opportunity to gain by investing in growth stocks.

An average return of growth stocks using time series modelling

Based on the possible ARIMA models of average return, the comparison is shown in Table 3. Herein, the assessment of AIC or BIC shows the model (5, 0, 1) and (1, 0, 1). However, in these models majority of AR and MA are insignificant i.e. 3 AR insignificant out of 5 lags. Thus, they cannot be considered the optimal model. As (1, 2, 1) and (3, 2, 1) have a majority of their coefficient as significant and less than 1, but among them, model (3, 2, 1) has the least AIC or BIC value. Thus, (3, 2, 1) is the optimal model for predicting an average return of growth stocks.

Results of all possible average return models
Table 3: Results of all possible average return models

The average return model based on the coefficient values is shown in below table

Average return forecasting model for growth stocks
Equation 2: Average return forecasting model for growth stocks

Where Y represents the average return of growth stocks and shows the error term of the model.

The selected model of average return predicts the AR values shown in Table 4. Herein, the annual average (AR) return shows that there has been a reduction in earning opportunities for the investors as the return value has decreased from -0.00156 to -0.00178. Though from FY 2022 there has been a slight improvement in the performance by rising in average return to -0.0018 but by FY 2024, again the growth stocks return has decreased. Further, an assessment of the earning opportunity for investors is shown in Figure 2.

Financial YearAnnual Average Return
2018-0.00179
2019-0.00156
2020-0.00178
2021-0.0023
2022-0.0018
2023-0.00164
2024-0.00213
Table 4: Annual return of growth stocks

Figure 2 below shows that there has not been much increase in the earning opportunities for the investors as witnessing the constant change, there is an absence of any trend in the dataset. Thus, despite the high average closing price after FY 2024, the earning opportunity for investors would not increase.

The predicted average return of growth stocks
Figure 2: The predicted average return of growth stocks

The average closing price of income stocks using ARIMA

As per the ARIMA models identified in the previous article the comparison of the models is shown in Table 5. Herein, the model (2, 2, 1) and (3, 2, 1) are the ones having the least AIC or BIC values. Herein, the model (2, 2, 1) has the majority of the AR and MA coefficients significant and lesser than 1. Thus, the optimal model for the average closing price of income stocks is (2, 2, 1).

Results of all possible Average closing price models using ARIMA
Table 5: Results of all possible Average closing price models

As per the coefficient values, the ARIMA model could be stated as

Average closing price forecasting model for income stocks
Equation 3: Average closing price forecasting model for income stocks

Where Y is the average closing price and represents the error term of the model.

Based on the model stated in equation 3, the prediction about the annual average closing price is made which is shown in Table 6. Herein, the value depicts that there has been a drastic fall in the average closing price of stocks post-2019 i.e. from 686.2885 in FY 2019 to 357.3055 in FY 2020. These values were further reduced to 157.4541 in FY 2021 but after FY 2022, the closing price improved to 246.1534. Though there has been slight performance improvement initially, since FY 2024, again, there has been a decline in the value of income stocks. Trend-based assessment of income stocks’ closing price is shown in Figure 3.

Financial YearAnnual Average Closing Price
2018843.416
2019686.2885
2020357.3055
2021157.4541
2022246.1534
2023342.6272
2024305.5574
Table 6: Annual closing price of income stocks

The below figure shows that there would be a sharp decline in the closing price of income stocks after FY 2019. Although by FY 2022, the performance would improve, the closing price would be low showing a presence of a downward trend, depicting a decline in earning opportunities for income stock investors.

Predicted closing price of income stocks using  ARIMA
Figure 3: Predicted closing price of income stocks

An average return of income stocks using ARIMA

Based on the possible models derived in the previous article the assessment of each of the models is shown in Table 7. Herein, model (3, 0, 16), (4, 0, 16) and (3, 0, 3) has the least AIC or BIC values.

Results of all possible average return models using ARIMA
Table 7: Results of all possible average return models

The ARIMA model for average return can be formulated as follows:

Average return forecasting model for income stocks
Equation 4: Average return forecasting model for income stocks

Wherein, Y represents the average return of income stocks while showing the error term of the model.

The prediction of the average return for income stocks based on the model defined in Equation 4 is shown in Table 8. With an annual average return value of -0.00421 for income stocks, there has been a rise in earning opportunities for the investors to -0.00182. But by FY 2022, the performance of stocks again declined. The return value could be decreased to -0.00174. Thus, there has been variation in the performance of income stocks but still, comparison with growth stocks shows that investors have more opportunity of earning due to higher return earning possibility in growth stocks.

Financial YearAnnual Average Return
2018-0.00224
2019-0.00421
2020-0.00182
2021-0.00159
2022-0.00174
2023-0.00152
2024-0.00249
Table 8: Annual return of income stocks

Assessment of annual average return shows that initially there has been a negative value of return but after FY 2024, investors would have the possibility of earning a better return. Initially, growth stocks show better-earning opportunities as most of the annual average return value of income stocks is below 0. But after FY 2024, income stocks would be a secure source of investment due to the presence of less variability below 0 compared to growth stocks wherein, huge fluctuations could be witnessed.

The predicted average return of income stocks using ARIMA
Figure 4: The predicted average return of income stocks

The average closing price of value stocks using ARIMA

Based on the identified possible ARIMA models in the previous article, the comparison of all the models is shown in the below table. Herein, the least value of AIC or BIC is for the model (1, 2, 1). With significant and less than 1 coefficient values for AR and MA, the model (1, 2, 1) is considered an optimal prediction model for the average closing price.

Results of all possible average closing price models using ARIMA
Table 9: Results of all possible average closing price models

As per the derived model of average closing price, the mathematical form of the model could be stated as

Average closing price forecasting model for value stocks
Equation 5: Average closing price forecasting model for value stocks

Wherein, Y shows the average closing price of the model while denoting the error term.

Further, the prediction of the average closing price values for value stocks is shown in Table 10 below. Herein, having a value of 132.919 in FY 2019, the closing price would witness a drastic decline in performance i.e. to 48.60683 in FY 2020. This declining performance for value stocks would improve in FY 2022. A detailed examination of the performance is shown in Figure 5.

Financial YearAnnual Average Closing Price
2018184.7788
2019132.919
202048.60683
202146.06175
202262.99034
2023121.2559
2024152.2917
Table 10: Annual closing price of value stocks

Trend shown in the below figure depicts that compared to the trend of growth and income stocks, value stocks would witness a rise in their performance by having a possibility of an upward trend. But despite this upward trend, the closing price value estimation would be less than the growth and income stocks value. Thus, value stocks investors due to the upward moving trend would have an opportunity of earning higher returns due to rising closing prices but this value would be less than income and growth stocks.

Predicted closing price of value stocks using ARIMA
Figure 5: Predicted closing price of value stocks

The average return of value stocks using time series modelling

The comparison of all identified ARIMA models for average return in the previous article is shown in Table 11. Herein, the assessment of AIC or BIC value shows that models (3, 0, 12) and (5, 0, 12) have the lowest values. However, for the model (3, 0, 12) majority of AR and MA coefficient values are insignificant and even more than 1. Thus, it would not be the optimal model for predicting the performance of average return. As, model (5, 0, 12) has a majority of coefficient values as more than 1 and significant, thus, (5, 0, 12) is the efficient and optimal model for average return.

Results of all possible average return models
Table 11: Results of all possible average return models

Based on the identified ARIMA model, the mathematical form of the model is stated below.

Average return forecasting model for value stocks
Equation 6: Average return forecasting model for value stocks

Where Y shows the average return of the model while depicting the error term.

The prediction of the annual average return is further shown in table 12 below. Herein, the value of the return is -0.0059 for FY 2019 which has decreased to -0.00248 in FY 2020 showing the availability of earning opportunities for the investors of value stocks. There has been a reduction in earning possibility for investors with the availability of decreasing possibility in future. As the value of the return is negative, there is the possibility of loss in future which is even higher than the income and growth stocks loss probability. A detailed assessment of the performance is shown in Figure 6 shown below.

Financial YearAnnual Average Return
2018-0.00357
2019-0.0059
2020-0.00248
2021-0.00256
2022-0.00302
2023-0.00301
2024-0.00291
Table 12: Annual return of value stocks

The below figure shows the prediction of average return movement for the next 10 years. Herein, in presence of high fluctuations, the return earning possibility is less as majorly the trend value is below 0. Comparing the performance with the income and growth stocks, the analysis shows that investors of value stocks have more probability of loss in the longer term.

The predicted average return of value stocks using ARIMA
Figure 6: The predicted average return of value stocks

Overall stocks’ performance in the Bombay Stock Exchange

To optimize returns on stock market investments, forecasting stock price trends is important. In this study, 303 stocks listed on the Bombay Stock Exchange for the period 2000-2020 were analysed. These stocks were split into three types:

  1. growth,
  2. income and,
  3. value stocks.

The prediction of stock performance in this study states that growth stocks offer a maximum earning possibility for investors. Although there is the presence of variation, the majority of the annual average return is above 0. Therefore, the investors would still benefit from this stock. Income stocks are secured investments, therefore provide less earning possibility in comparison to growth stocks. However, the amount of inherent risk is also less. Thus, there is a huge variation in the average return for income stocks at the level above and below 0 value. However, in the longer term due to the upward focused trend, there is the possibility of a safe return for investors.

Value stocks are a source of risky investment. With the majority of annual average return values below 0 levels, the investors of these stocks would bear the loss and this possibility rises in the longer term as the trend is focused on negative values. Hence, investors seeking more returns should invest in growth stocks, but investors who prefer more secure investments should opt for income stocks.

References

  • Jampala, R. C., Goda, P. K., & Dokku, S. R. (2019). Predictive analytics in stock markets with special reference to BSE Sensex. International Journal of Innovative Technology and Exploring Engineering, 8(6), 615–619. https://doi.org/10.35940/ijitee.F1127.0486S419
  • Perwej, Y., & Perwej, A. (2012). Prediction of the Bombay Stock Exchange (BSE) Market Returns Using Artificial Neural Network and Genetic Algorithm. Journal of Intelligent Learning Systems and Applications, 04(02), 108–119. https://doi.org/10.4236/jilsa.2012.42010
NOTES

Discuss