Already, some data collection—such as web-scraping prices—is executed with Python. Python also seems likely to increasingly handle organizing that data. In time, this could extend to the actual modeling process. There’s always resistance to this type of change, but there are push and pull factors.
The pull factor: as "passive" strategies develop, the dividing line between passive and active will shift, requiring active approaches to tap a broader set of data in order to achieve idiosyncratic returns and stay ahead of the competition. The push factor is cost: it’s likely more efficient in time and personnel to manipulate data in Python, and inexorably declining fees and pressure on margins will force the transition.
This process has advanced the most in quantitative modeling. Modeling for truly systematic investment wasn’t taking place in Excel in the first place, but here the methodology is changing by adopting machine learning—and in some cases the claim of artificial intelligence. The push and pull factors are similar, but the jury remains out on how far the process will play out. Adoption of machine learning for manipulating and extracting data seems set to grow dramatically, but it’s unclear to what extent it can be applied to making actual investment decisions.
There are also open questions on how much complexity is acceptable in financial models, especially when they fail; preferences may develop for different kinds of machine learning models. So-called ensemble models like random forests can be constrained structurally so that they can be mapped onto the "real world," while neural nets and support-vector machines lack that option.
Factors Could Become the New Asset Classes
The past four decades have been an extraordinary time, with prices on financial assets rising and diversification among them plentiful. Now, investors face a strategic valuation problem. This isn’t necessarily bearish, but it does imply low expected real returns on major asset classes. And if inflation rises, bond diversification may not be as effective, which could increase portfolio risk.
Set against this backdrop, valuation spreads within asset classes are very extreme. At the same time, we can demonstrate that the diversification benefits of factors tend to be more stable than those of asset classes. Perhaps at least part of the answer to the current risk/return dilemma is to consider asset classes and factors as interchangeable. After all, securities markets are ultimately merely composed of assets issued by companies and governments—who’s to say that a more "primal" way of dividing them up is by their legal structure (asset class) rather than financial characteristics (factors)?
One pushback on this notion is that it requires investors to deploy assets in a way that, for the past decade, would have been very suboptimal.
For 10 years, the cheap and accepted option of simply buying passive stocks and bonds has been a great trade. Meanwhile, the average return of factors has been below their longer-run average rate of return, with the case of apparent "failure" of the value factor prominent in that. However, we think that cyclical factors explain at least some of this. A factor approach would certainly not be a panacea, but we think it might be essential to achieving a given level of risk/return and will likely appeal more to investors who believe that the post-pandemic world will be fundamentally different.
Organizational Implications: Falling Silos?
What are some of the organizational ramifications of the innovations we’ve discussed? In an environment in which it’s harder to generate a given level of real return, it’s more likely that the industry will need to consider its output as a "return stream" rather than a certain kind of fund "product."
This could lead to realigning organizations based on the nature or characteristics of return streams (alpha, factors, security-specific, macro) rather than asset class. Moreover, the switch from Excel to Python for financial models seems poised to blur the distinction between quantitative and "fundamental" models—and hence the historic partition between these modes of investing.
At the same time, to the extent that this modeling change enables the use of (expensive) new data sets, it implies that the new models that are needed, and the teams that develop them, become functions spread across organizations rather than limited to a specific asset class. That shift could make new modeling capabilities more cost-efficient. It might be necessary to break up long-established investment industry silos before these changes can be fully realized.