![Will generative AI generate returns? [PODCAST]](/content/dam/thrivent/fp/fp-insights/advisors-market360-podcast/compass-insights-card-v1.jpg/_jcr_content/renditions/cq5dam.web.1280.1280.jpeg)
Will generative AI generate returns? [PODCAST]
The opportunities and pitfalls of investing in AI-related companies.
The opportunities and pitfalls of investing in AI-related companies.
08/27/2024
FUND COMMENTARY
One of the keys to the investment success of Thrivent’s mutual funds is a portfolio management process we refer to as “systematic alpha.” It guides decisions in building and managing many of our portfolios.
By understanding systematic alpha, you can gain a better understanding of how we manage some of our equity portfolios at Thrivent Asset Management. The Thrivent Systematic Alpha Team includes six portfolio managers and four quantitative research analysts. As the assets under management in the funds that utilize the systematic alpha approach have continued to grow, we have added more research capacity to the team.
For each strategy we manage, our goal is to build a portfolio of securities that generates excess returns, or alpha, relative to the stated benchmark—regardless of the market environment. It’s also important that each portfolio remains true to its investment style to avoid unintended risks.
We do this by using a systematic, quantitative approach for three aspects of investment management—security selection, portfolio construction and risk management.
This time-tested process is defined, repeatable and consistent for any universe of securities that we manage. But while the overarching process is the same for all products that we oversee, different markets or asset classes are managed using different proprietary sets of factors unique to each strategy. For example, the factors used for U.S. stocks may be different than the ones used for Japanese stocks.
We tend to have much smaller positions and more diversified portfolios than managers who use fundamental analysis. While fundamental managers make active bets on individual companies and try to maximize the idiosyncratic alpha in their portfolios, we prefer to diversify away a lot of the individual company risks and concentrate the alpha potential of the portfolio on a range of factors that have been shown to be predictors of stock performance.
For each portfolio we manage, we start by defining the universe of investable securities, based on region, market capitalization and style of the benchmark. For example, for a domestic small-cap value strategy, we run screens to narrow our list down to U.S.-only stocks and weed out the large-sized companies.
Then we use quantitative inputs from our risk model to assess the growth and value of the securities in the universe to ensure style consistency. We follow that up by establishing a customized model of 60 to 80 factors to be used in the management of each portfolio. We derive these factors from six overarching themes that apply to all portfolios:
While the six themes are universal across all portfolios we manage, the underlying factors within these themes vary depending on the region, country and industry. Each of the 60 to 80 factors represents a characteristic of a stock that can be quantified (such as price-to-earnings ratio) and has predictive power for future returns or volatility. Factors may encompass a stock’s valuation, risk, historical growth, quality, price history, or sensitivity to macroeconomic indicators.
Once the model is established for a portfolio, we re-evaluate and optimize it on a monthly basis to ensure we are using the best set of factors to achieve each portfolio’s goals. Typically, only one or two factors change monthly, so it’s more of an evolutionary process over time.
An integral part of our systematic alpha approach involves regular rebalancing of our portfolios, while simultaneously managing risk through diversification. While diversification can help reduce market risk, it does not eliminate it.
We believe our disciplined method takes the emotion out of security selection and the portfolio construction process and leads to a lower risk of style drift within the portfolio. The approach also results in less chance of an overconcentrated portfolio, either by an individual holding, sector or country. Our portfolios typically range between 150 and 250 securities.
Because the approach is more diversified, it diminishes the portfolio’s exposure to idiosyncratic risk, which is the risk of a sharp price decline of one stock, or even a certain industry or sector due to a specific event that doesn’t impact the overall market.
We evaluate every security in our investable universe on a daily basis using the set of factors in the model and assign a score to each security, which represents its expected alpha. Then, typically bi-weekly, we will rebalance each of our portfolios based on these scores in order to maximize expected returns while keeping portfolios in line with the tracking error.
As a result, our portfolios typically experience approximately 3% turnover every other week to incrementally keep the portfolio in line with how the securities are scoring in the model. That typically equates to 75% to 80% annual turnover.
We believe Thrivent’s systematic alpha approach is unique in the way it integrates machine learning into the monthly factor selection and weighting process, instead of using a static linear model.
We rely on an iterative technique called adaptive boosting, which is typically used in “Big Data” and artificial Intelligence (AI) systems, to help calculate and correct prediction errors. Once a month, we run this process on historical data from our pooI of factors to identify which combination of factors and weights has been most predictive of future returns.
By iterating through this process, we obtain an optimized set of factors and their weights, effectively training the model to correct its previous forecasting errors.
Systematic alpha is a quantitative-driven process, and most of the time we follow the trade recommendations generated by the optimizer. That said, our portfolio management team does manually review the entire list of recommended stocks each week to scan for issues that can’t be quantified in the model but could convince us to avoid the security.
Some examples would include recent company news or merger and acquisition activity that’s not yet reflected in the data. In those cases, we may choose to rerun the optimization with that security excluded and replace it with another option.
Our goal is to continually build out and enhance our alpha models by developing more proprietary tools that are customized to our investment process.