The move from active to passive investing has been a secular theme for many years, and it has disrupted the asset management industry. As active fund managers face increasing pressure to outperform their benchmarks, the hunt for alpha is becoming a more difficult exercise. Some asset managers prefer to use quantitative investing techniques to construct portfolios and use machines to identify how the portfolio should behave in all facets. This can include how new investment ideas are generated, how much should be bought or sold and when changes to the portfolio should occur. This type of investing has coined the term “rise of the machines,” as more and more assets under management move toward computer-based investing. This was most evident when Blackrock, the largest asset manager in the world, decided to move a portion of the money they manage into a quant-based strategy unit called “systematic active equities.”
There is certainly merit in utilizing quant based methods for investing as they can provide two distinct advantages. Using machines allows investment firms to scale their research process and puts no limits on the number of securities that can be analyzed. The second benefit is that machine-based investing reduces the behavioral biases that humans exhibit (i.e. loss-aversion, confirmation bias).
What becomes interesting is how we can use machines and data to produce valuable insights that were otherwise difficult to observe, or more importantly, discover new insights using non-traditional data sources. The term “big data” has gained significant attention over the years, as the velocity and volume of data available today is growing at an exponential rate. Consider these examples: using satellite images of a parking lot to better understand customer traffic at a grocery store, sensors to track soil moisture for farmers, or credit card transactions to look at retail sales.
While there are exciting developments in big data, machine learning and artificial intelligence, we need to be mindful of the many challenges big data faces, including privacy, regulation, cost, and data availability. This will be studied over the years to determine how and if alternative data sets can be married into the investment process efficiently while being able to generate a sustainable source of alpha. While the future cannot be predicted, one thing is for certain – fund managers will look to use data-driven techniques to complement the investment process in a term I like to call “man plus machine.”
Looking at the StarMine Text Mining credit model, we can see how machines directly improve something most humans do on a daily basis – reading. The Text Mining model is trained to detect bullish or bearish sentiment by employing machine learning techniques to scan for words and phrases contained in news stories, company filings, transcripts, and broker research. Once this is done, the model produces a 1-100 ranking to inform the practitioner whether a company looks sound from a credit default perspective.
As seen in Exhibit 1, Sears Holding (SHLD.O) had a ranking of 1 which implied the company was in financial trouble. This is also confirmed when looking at the probability of default over a 12 month period, which stood at 41.3%.
Exhibit 1: StarMine Text Mining Credit Model for Sears Holding
It is important to observe that our model implied rating (CC) was indicative of future credit rating revisions by S&P. We can see that in January 2018, S&P eventually downgraded Sears to CC, a rating that was forecasted almost two years ago by our StarMine implied rating.
It is beneficial for the analyst or portfolio manager to view StarMine model scores at a company level to help make better investment decisions. But before analysts or portfolio managers decide which stock to put into their portfolio, they first need to define the investment philosophy of the fund: Growth-oriented? Value-oriented? Or a combination of different styles (i.e. a multi-factor portfolio)?
To help with this complex task, we can utilize a back-testing tool which can test your investment strategy to see how it would perform over a defined time period. Using Thomson Reuters QA Point can help us accomplish this very easily in a point-and-click web-based environment.
In the example below, we will create a multi-factor model containing a selection of StarMine models to see how it will perform against the S&P 500, arguably the most difficult benchmark to beat. Exhibit 2 shows how we can create our factor-based portfolio, which includes our own user-defined weights. It is worth noting that we would expect to see stronger results when running a strategy in a less efficient market (i.e. emerging markets).
Exhibit 2: Creating a Multi-Factor Model in QA Point
Once we have created our portfolio, we can now use the Factor Tester to see how our investment strategy will perform. During this process, we can define the universe where we want to apply our strategy — in addition to picking our benchmark. We will conduct our test over a 20-year time period and display the data in a decile format. We can see the results of our back-test in Exhibit 3.
Exhibit 3: Performance of User-created Model vs. Benchmark
Looking at the output, we can see that our multi-factor strategy has significantly outperformed the benchmark, as seen by the green line: Fractile 1, which buys the top 10 percent of companies with the strongest StarMine model scores. Fractile 1 has a cumulative 20-year return of 2,053% compared to the benchmark return of 249% (returns are on a total-return basis). We can also see the top bottom statistics over to the right, which shows how our strategy performed if we pursue a long/short strategy – buying the companies with the highest StarMine model ranking, and selling the worst ranked companies. This strategy would deliver an impressive 9.1% annualized return over 20 years – an impressive track-record which would not many can claim these days.
To wrap up, I believe humans will not be overtaken by machines, but combining the best attributes of both can deliver a robust investment process and help find that needle in a haystack, since the haystack seems to be getting smaller and smaller.