October 17, 2023

Risky Business And The Art Of The Modern Model Portfolio

Risky Business And The Art Of The Modern Model Portfolio

Imagine the founder of a startup who with each round of funding decides to go all in with all of his capital. This strategy leads to one of two outcomes: either his startup fails or it achieves a tenfold increase in value, each with a 50/50 chance of occurring. 

Mathematically this is expressed as Expected Outcome = 0.5 × 0 + 0.5 × 10 = 5. So far so good.

But your experience as an investor is very different from the experience of a founder. 

The founder encounters what's called the time-average effect: over a long enough time period his hit or miss approach will lead to certain failure. If he keeps flipping coin then eventually he will get a tails.

As an investor, on the other hand, you experience what's called the ensemble-average: with a large enough portfolio of companies you can anticipate, on average, a fivefold increase in your portfolio in each funding cycle.

This concept is fundamental to investment management. It's why managers seek diversification - when you diversify you're seeking out the ensemble-average. It's why listed companies blow up all the time but equities indices keep rising. It applies, explicitly or implicitly, in portfolio theory, discounted cash flow models, aggregate indicator models and any number of foundation models in finance and investing. 

There's Nothing More Practical Than A Good Theory

But in reality the world is much more nuanced. For the startup founder, a more conservative plan that 5x's or 0.5x's his company's value has a geometric expected outcome of > 1 and increases 58% every year. Thoughtfully creating value outperforms reckless risk taking in the long run.

Similarly, the ensemble average for investors only trends upwards if there are an infinite number of startups in your portfolio. Otherwise theoretically your fund will blow up eventually too.

This is why in markets like alternative assets, where price discovery is especially hard, investors claim that their screening process is more art than science. Venture capitalists specialize in a specific sector for instance, and weight features like the management team’s track record in their binary classification of whether to invest or not. 

Even the most robust theories are a framework for understanding how the world works - they're not a guarantee of success - and arguably stock selection really is more art than science under certain circumstances. Machine learning models offer us a way to integrate this investor intuition into the decision making process while maintaining a framework that is scientific enough to scale.

Here’s how they work in a nutshell.

Utilizing Machine Learning for Portfolio Optimization

  1. Feature Identification: The first step is to identify key features that influence a startup's success. The way you define these features is your own secret sauce as an investor and includes market trends, financial health, team experience etc.. whatever proprietary models you’re using already. 
  1. Data Collection: The machine learning model requires data on these features. This data could be historical performance metrics of startups, market analysis reports, or financial statements. This is a data engineering exercise and requires understanding how much of what kind of data you need, how often and when.
  1. Model Training: With this data, the machine learning model is trained to recognize patterns and correlations between these features and the startups' outcomes (success or failure)
  1. Predictive Analysis: Once trained, the model can predict the potential success of new startups based on their features. It can assess risks and likely returns, helping managers make informed decisions about which startups to include in their portfolio.
  1. Diversification Strategy: The model can suggest a diversification strategy by selecting a mix of startups with varied features. This reduces the risk that comes from concentrating investments in startups with similar characteristics or in the same market sector.
  1. Continuous Learning and Adjustment: As the model gets exposed to more data over time, its predictions can become more accurate. This allows for ongoing refinement of the investment portfolio, adapting to new trends and market changes.

Again, using AI doesn't give us any super intelligence or any guarantee of success. It does help us create an approximation function of the real world and in doing so lets us integrate art and science. Applying more scientific methods to your business helps you consistently make better decisions and gives you a considerable edge over your competition. 

By its nature, integrating your proprietary secret sauce into a scientific framework also allows you to test hypotheses and measure results. This use of statistical and probabilistic methods to assess and manage risks promotes more stable and sustainable growth.

Work Smarter And Harder

We tend to get excited about AI here, and we’re biased in favor of accelerating its application to finance. Obvs. But using technology to get a more nuanced, data-driven approach, is especially effective in complex hard to price markets where "art" differentiates good investments from bad ones. And there are huge benefits to the wider economy in this.

In this sense AI is not merely enhancing investment management; it’s redefining it. By enabling better investment decisions it’s facilitating a far more effective allocation of capital, and unlocking potentially vast reservoirs of value that may otherwise remain confined by the constraints of the conventional system. Who wouldn’t get excited about that?