Abstract
We present a model based on trend filtering and regularization for performing factor analysis. Furthermore, the trend filtering method proposed in our model allows incorporating views represented as scenarios. Therefore, factor models can be optimized to explain not only trends of a given data series but also trends reflecting outlooks. As an application to finance, the proposed model constructs a sparse factor model on the trend of an index. Factor analyses on trend-filtered series provide factor models for describing trends, which are valuable for modeling long-term horizons. We also analyze the effect of trends on factor model selection in the US stock market. Further analyses include an investigation of factors during crisis periods and a comparison when various scenarios are considered.
Similar content being viewed by others
Change history
03 April 2021
The article has been updated due to the word “oneword” erroneously used instead of “dataset” present in line 387 of this article.
Notes
Empirical analyses in this section are performed using Matlab R2016b (64-bit) and optimization is solved with CVX Version 2.1.
Factors (1)–(11) in Table 1 are retrieved from the Federal Reserve Economic Data (FRED), and industry data along with US stock market returns are collected from the data library of Kenneth R. French. Industry factors are included to represent various sectors of the stock market (Cavaglia et al., 2000).
Note that, while key indicators are included in the list, this is not a comprehensive list or the ideal combination of market factors.
While there is no general guideline for selecting the value of γ, Yamada and Yoon (2016) offer insight on selecting the tuning parameter based on the relation between H–P and \(\ell_{1}\) trend filtering methods.
References
Ang, A. (2014). Asset management: A systematic approach to factor investing. Oxford University Press.
Banz, R. W. (1981). The relationship between return and market value of common stocks. Journal of Financial Economics, 9(1), 3–18.
Baxter, M., & King, R. G. (1999). Measuring business cycles: Approximate band-pass filters for economic time series. Review of Economics and Statistics, 81(4), 575–593.
Bertocchi, M., Consigli, G., & Dempster, M. A. H. (2011). Stochastic optimization methods in finance and energy: New financial products and energy market strategies. Springer.
Brodie, J., Daubechies, I., De Mol, C., Giannone, D., & Loris, I. (2009). Sparse and stable Markowitz portfolios. Proceedings of the National Academy of Sciences, 106(30), 12267–12272.
Carhart, M. M. (1997). On persistence in mutual fund performance. Journal of Finance, 52(1), 57–82.
Cavaglia, S., Brightman, C., & Aked, M. (2000). The increasing importance of industry factors. Financial Analysts Journal, 56(5), 41–54.
Cerniglia, J., & Fabozzi, F. J. (2018). Academic, practitioner, and investor perspectives on factor investing. Journal of Portfolio Management, 44(4), 10–16.
Chen, J., Dai, G., & Zhang, N. (2020). An application of sparse-group lasso regularization to equity portfolio optimization and sector selection. Annals of Operations Research, 284(1), 243–262.
Consiglio, A., Cocco, F., & Zenios, S. A. (2007). Scenario optimization asset and liability modelling for individual investors. Annals of Operations Research, 152, 167–191.
Dimson, E., Marsh, P., & Staunton, M. (2017). Factor-based investing: The long-term evidence. Journal of Portfolio Management, 4(15), 15–37.
Dopfel, F. E., & Lester, A. (2018). Optimal blending of smart beta and multifactor portfolios. Journal of Portfolio Management, 44(4), 93–105.
Dupačová, J., Consigli, G., & Wallace, S. W. (2000). Scenarios for multistage stochastic programs. Annals of Operations Research, 100(1–4), 25–53.
Fama, E. F., & French, K. R. (1992). The cross-section of expected stock returns. Journal of Finance, 47(2), 426–465.
Fama, E. F., & French, K. R. (1993). Common risk factors in the returns on stocks and bonds. Journal of Financial Economics, 33(1), 3–56.
Fama, E. F., & French, K. R. (2015). A five-factor asset pricing model. Journal of Financial Economics, 116(1), 1–22.
Fama, E. F., & MacBeth, J. D. (1973). Risk, return, and equilibrium: Empirical tests. Journal of Political Economy, 81(3), 607–636.
Hodrick, R. J., & Prescott, E. C. (1997). Postwar U.S. business cycles: An empirical investigation. Journal of Money, Credit, and Banking, 29(1), 1–16.
Ho, M., Sun, Z., & Xin, J. (2015). Weighted elastic net penalized mean-variance portfolio design and computation. SIAM Journal on Financial Mathematics, 6(1), 1220–1244.
Kim, J. H., Kim, W. C., & Fabozzi, F. J. (2017). Robust factor-based investing. Journal of Portfolio Management, 43(5), 157–164.
Kim, S. J., Koh, K., Boyd, S., & Gorinevsky, D. (2009). L1 trend filtering. SIAM Review, 51(2), 339–360.
Kim, W. C., Kwon, D. G., Lee, Y., Kim, J. H., & Lin, C. (2020). Personalized goal-based investing via multi-stage stochastic goal programming. Quantitative Finance, 20(3), 515–526.
Kim, W. C., & Mulvey, J. M. (2009). Evaluating style investment—Does a fund market defined along equity styles add value? Quantitative Finance, 9(6), 637–651.
Köksalan, M., & Şakar, C. T. (2016). An interactive approach to stochastic programming-based portfolio optimization. Annals of Operations Research, 245(1–2), 47–66.
Kremer, P. J., Talmaciu, A., & Paterlini, S. (2018). Risk minimization in multi-factor portfolios: What is the best strategy? Annals of Operations Research, 266(1–2), 255–291.
Madhavan, A., Sobczyk, A., & Ang, A. (2018). What’s in your benchmark? A factor analysis of major market indexes. Journal of Portfolio Management, 44(4), 46–59.
Melas, D., Suryanarayanan, R., & Cavaglia, S. (2010). Efficient replication of factor returns: Theory and applications. Journal of Portfolio Management, 36(2), 39–51.
Mulvey, J. M., Hao, H., & Li, N. (2018). Machine learning, economic regimes and portfolio optimisation. International Journal of Financial Engineering and Risk Management, 2(4), 260–282.
Mulvey, J. M., & Liu, H. (2016). Identifying economic regimes: Reducing downside risks for university endowments and foundations. Journal of Portfolio Management, 43(1), 100–108.
Pedersen, A. M. B., Weissensteiner, A., & Poulsen, R. (2013). Financial planning for young households. Annals of Operations Research, 205(1), 55–76.
Ramdas, A., & Tibshirani, R. J. (2016). Fast and flexible ADMM algorithms for trend filtering. Journal of Computational and Graphical Statistics, 25(3), 839–858.
Roll, R., & Srivastava, A. (2018). Mimicking portfolios. Journal of Portfolio Management, 44(5), 21–35.
Tibshirani, R. J. (2014). Adaptive piecewise polynomial estimation via trend filtering. Annals of Statistics, 42(1), 285–323.
Vidal-García, J., Vidal, M., Boubaker, S., & Manita, R. (2019). Idiosyncratic risk and mutual fund performance. Annals of Operations Research, 281(1–2), 349–372.
Yamada, H. (2017). Estimating the trend in US real GDP using the \(\ell_{1}\) trend filtering. Applied Economics Letters, 24(10), 713–716.
Yamada, H., & Jin, L. (2013). Japan’s output gap estimation and \(\ell_{1}\) trend filtering. Empirical Economics, 45(1), 81–88.
Yamada, H., & Yoon, G. (2016). Selecting the tuning parameter of the \(\ell_{1}\) trend filter. Studies in Nonlinear Dynamics and Econometrics, 20(1), 97–105.
Acknowledgements
This research was supported by Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (NRF-2018R1C1B6004271).
Author information
Authors and Affiliations
Corresponding author
Additional information
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Appendix
Appendix
1.1 Appendix 1. Property of trend filtering with scenarios
Property Suppose trend filtering results in trend \(z \in {\mathbb{R}}^{n}\) for a given data \(y \in {\mathbb{R}}^{n}\). Then, consider appending scenarios to the given data y. Trend filtering for y with scenarios will result in a trend that is different from z if the mean of the scenario values lies on the line that extends the final segment of the original trend z.
Proof We show the case for when the second-order difference matrix, D(2), is used for trend filtering and scenarios are added for a single period, but the argument can be extended to more general cases.
Suppose in addition to n observed data points, N scenarios for a single period is considered, which are denoted as
Then, the length of the trend will be extended one period due to added scenarios for a single period (i.e., \(z \in {\mathbb{R}}^{n}\) becomes \(z \in {\mathbb{R}}^{n + 1}\) where \(z_{n + 1}\) reflects the added period), and the objective function will be extended to also include
If the trend z were not changed and the final line segment of the trend is extended one period, the following must hold,
Then, for any other \(z_{n + 1}^{0}\), it must be that
For large N, the right-hand side is minimized for \(z_{n + 1}^{0} \approx \frac{{\sum s_{i} }}{N}\).
Thus, the trend (at least the final segment of the trend) will be altered in general unless \(z_{n + 1}^{*}\) is purposely set to \(z_{n + 1}^{*} = 2z_{n} - z_{n - 1} \ne \frac{{\sum s_{i} }}{N}\).
1.2 Appendix 2. Two-step procedure for analyzing factors
The formulation given by (8) can be performed in two steps by optimizing the trend and then finding the factor model. For simplicity, we show that formulations (5) and (6) can be solved through (9) and (10).
In formulation (5), small values of δ relative to γ will result in factors having minimal influence on the trend filtering part of the objective function. Therefore, the two-step procedure has the effect of solving the combined problem with a small value of δ. In other words, if we define \(\beta_{0}\) as the factor loadings from solving (10), then factor loadings from solving (5) with small values of δ should be the same as \(\beta_{0}\). This is clearly shown empirically in Fig.
6, which plots the L2 distance between the factor loadings from the two-step procedure, \(\beta_{0}\), and the factor loadings from the original formulation with various values of δ. When δ is set to values smaller than 0.01, the distance becomes less than 0.0003 and the curve flattens. This is the case when the same dataset (1970–2018) from Sect. 4 is used, the value of γ is set to 0.5, and without regularization. Since the primary goal of the analyses we perform in Sect. 4 is to find a factor model of a trend, we use the two-step procedure to find the factor model after filtering the trend.
1.3 Appendix 3. Description of factors
Table
8 includes description of the industry factors, which are retrieved from the data library of Kenneth R. French. Table
9 presents summary statistics and Table
10 shows correlation of all factors that are used in the analyses from 1970 to 2018.
Rights and permissions
About this article
Cite this article
Kim, J.H., Kim, W.C. & Fabozzi, F.J. Sparse factor model based on trend filtering. Ann Oper Res 306, 321–342 (2021). https://doi.org/10.1007/s10479-021-04029-9
Accepted:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10479-021-04029-9