The investment industry is undergoing a change, largely because of advances in technology. Investment professionals are integrating latest technologies corresponding to machine learning (ML) throughout the investment process, including portfolio construction. Many asset managers are starting to include ML algorithms into the portfolio optimization process to create more efficient portfolios than can be possible using traditional methods corresponding to mean-variance optimization (MVO). These trends require a fresh take a look at how ML is changing the portfolio construction process.
Investors profit from a basic understanding of ML algorithms and the impact of those algorithms on their portfolios. Ultimately, the strategies that asset managers use to construct client portfolios have a direct impact on the tip investor. Investors should subsequently be sufficiently aware of those methods as they have gotten increasingly popular. This article goals to supply an outline of the role that ML algorithms play within the portfolio optimization process.
background
The term “machine learning” was first utilized by AL Samuel in 1959. Samuel introduced a experiment by teaching a pc to play checkers and concluded that the pc had significant learning potential. These results paved the best way for further research on the topic and led to the event of increasingly powerful and complicated ML algorithms within the many years that followed. As a result, many industries, including investment management, have adopted these technologies lately.
ML algorithms are particularly useful in relation to analyzing high-dimensional data or data sets with nonlinear relationships, which is becoming increasingly common with the arrival of unstructured data and other alternative data sources. The two essential categories for ML are supervised learning and unsupervised learning. In supervised learning, the ML algorithm detects patterns between a set of features (i.e., input variables) and a known goal variable (i.e., output variable).[1]. This is named a labeled dataset since the goal variable is defined. However, in unsupervised learning, the dataset is unlabeled and the goal variable is unknown, so the algorithm tries to detect patterns within the input data. Annex 1 describes a number of the common ML algorithms currently utilized by investment professionals.
Figure 1: Common machine learning algorithms in investment management.
| ML algorithm | Description |
| Smallest absolute shrinkage and selection operator (LASSO) | A type of penalized regression that features a penalty term for every additional feature included within the regression model. The goal of this regularization technique is to create a parsimonious regression model by minimizing the variety of features and increasing the accuracy of the model. |
| -Means clustering | Divides data into clusters. Each remark in a cluster must have similar characteristics to the opposite observations, and every cluster ought to be clearly different from the opposite clusters. |
| Hierarchical clustering | Two types: bottom-up hierarchical clustering, which aggregates data into progressively larger clusters, and top-down hierarchical clustering, which breaks data into progressively smaller clusters. This results in alternative routes of grouping data. |
| Artificial neural networks (ANN) | A network of nodes that accommodates an input layer, a hidden layer, and an output layer. The input layer represents the features, and the hidden layer is where the algorithm learns and processes the inputs to generate the output(s). These algorithms have many uses, including speech and facial recognition. |
Figure 2. Factors expected to steer to significant changes in job profiles over the subsequent 5 to 10 years.
Portfolio optimization
The development of neural networks within the Sixties laid the inspiration for lots of the alternative methods for portfolio optimization using ML. In addition, the emergence of “expert systems”[2] has led investment professionals to increasingly depend on machines to resolve complex problems. Some of the primary applications of expert systems in finance are Trade And Financial planning Expert systems.
The use of ML algorithms within the portfolio construction process has gained popularity lately as investment professionals look for added ways to extend portfolio returns and gain a competitive advantage. In particular, integrating ML algorithms into the portfolio construction process can overcome the challenges and limitations of traditional portfolio optimization methods corresponding to MVO.
A key limitation of MVO is that it only considers the mean and variance of returns when optimizing a portfolio and doesn’t take note of the skewness of returns. However, in point of fact, investment returns are likely to exhibit skewness. In particular, Research has shown that growth stocks, on average, have higher positive skewness of their returns than value stocks. To account for the potential non-normality in investment returns, some investment professionals have chosen to construct portfolios using mean-variance-skewness optimization models and even mean-variance-skewness-kurtosis optimization models. However, these models result in multi-objective optimization problems. ANNs can efficiently construct mean-variance-skewness optimal portfolios to Fix this limitation.
Another drawback of the MVO is that it prevents investors from expressing their views on the longer term performance of assets. For example, an investor might expect bonds to perform higher than stocks over the subsequent six months. The Black-Litterman model (1992) allows investors to include these perspectives into the technique of portfolio optimization. A Alternative approach consists in integrating the Black-Litterman model (1992) with neural networks, which allows for top returns in comparison with the benchmark without taking excessive risks.
The inputs in MVO are sensitive to measurement error, which is very true for the estimate of expected return. Therefore, MVO has the potential to create “optimal” portfolios that perform poorly. Reverse optimization generally is a useful alternative to develop more accurate estimates of expected return. Investment professionals can then use these improved estimates as inputs in traditional MVO to generate more efficient asset allocations. Investment professionals may use ML algorithms to Predicting stock returns and incorporate these estimates into MVO. Alternatively, a recent study developed a Improved portfolio optimization Approach that consists of using a correlation shrinkage parameter to enhance the estimated Sharpe ratios after which constructing optimal portfolios based on these estimates.
Finally, estimating the covariance matrix is a significant challenge in portfolio optimization, especially for high-dimensional data. LASSO models can address this challenge by providing more accurate estimates of the covariance matrix than conventional methods, which is a vital input for MVO.
Conclusions
How do these trends impact investment professionals? Clearly, the investment industry is evolving rapidly in response to latest technologies. Investment professionals expect latest analytics methods corresponding to ML to significantly change job profiles in the approaching years. As a result, practitioners are starting to integrate ML algorithms into all areas of the investment process.
Many asset managers try to realize a competitive advantage by creating portfolios with higher returns at a given level of risk (i.e., higher Sharpe ratios) by integrating ML algorithms into the portfolio optimization process. In addition, ML algorithms can overcome lots of the challenges and limitations of traditional portfolio optimization methods, which has led investment professionals to hunt more efficient methods of portfolio construction. Investors will profit from increased awareness of those trends to higher understand the impact of latest optimization methods on their portfolios.
[1] In some cases, the dataset can have multiple goal variable.
[2] An expert system is a pc program that may solve a posh problem that is generally solved by human experts. See: Expert system | AI, knowledge representation and argumentation | Britannica
