You’re a finance graduate, fresh out of school and you get hired by one of the big finance shops; an investment bank, less so a private equity firm, and recently more likely, a venture capital firm. You don’t get thrown into building financial models for live deals straight away, you’re rather tasked with the mundane, loosely structured task of research. That’s not the highwire act of testing a hypothesis and arriving at a sound thesis. Nope, you just log on to the interwebs and collect all the data you can find to pass on to an associate to force-feed their due diligence material on a certain opportunity they’re evaluating.Analysts do top-down research, which involves scouring the web for industry reports for tidbits of insight and data. They also do bottom-up research, which is the equivalent on logging on whatever platform your employer is subscribed to (for most sell-side analysts, that’s Bloomberg, and for buy-side ones, that’s whatever platform holds private data) and gathering sparsely-populated competitor’s financial and operational data in hopes of piecing together a narrative.
One of the first things I learned in my undergraduate finance classes is that finance is more an art than a science, which is code for, "The numbers can say anything you want them to". And because art is subjective, an analyst’s primary task is to weave those numbers into a cohesive narrative they can sell.
Analysts pride themselves on their technical prowess and their Excel wrangling skills. They remember the hotkeys on the keyboard, can create the most beautiful charts, and build the most complex financial models. The reality is everyone’s working with the same data, superimposing their own biased, ungrounded narratives onto them in hopes of putting together a distinct insight they can share. I think there’s a better way.
The overfitting problem
You’re analyzing this company, combing over its financial performance, management’s forecasts and the investor presentations. Let’s call that company Databricks & Mortar, a data infrastructure and analytics platform. You believe competition is rough and no solution can disrupt the three incumbent cloud providers. That position turns into your bias and you go about your research scouring the web for data points to confirm that point of view; everything from market maps showing that concentration, to customer surveys showing high retention, high switchover costs. You do the google searches and skim through the Mckinsey and Deloitte reports on the “State of cloud infrastructure 2024”, spread them across 50 tabs and skim through the paragraphs looking for figures to validate your hypothesis “oh, Databricks & Mortar is dead within 3 years” you think to yourself. What you would be doing there is a classic case of what they call a confirmation bias. Sure you collected that data, knitted them together into a solid “data-driven” narrative about the company’s future and it would probably stand scrutiny from up the chain.
Databricks & Mortar recently went public and you’re initiating coverage on the stock. Your narrative makes its way to your model, you use your intricate formulas and granular metrics to drive your forecasting. Your model says the share is worth $5 and at the time the stock was trading at $10. One year later the price jumps to $25, but you hold your ground thinking the incumbents will crush it, the stock is trading on future expectations it will never grow into and you even get on calls with investors citing cases of other companies in the cloud space that crashed and burned to support your narrative. Another year goes by, Databricks & Mortar is now profitable and revenue growth remains in the double digits. You clearly missed something, you never expected an up and coming solution to go up against incumbents. You feel blindsided, but you get over it and remember that you’re only ever going to be right 51% of the time.
Anecdotal evidence is good
Whichever part of the spectrum you land in finance, you’re taught the importance of data to craft narratives and draw conclusions. The higher you are on that spectrum -That is the closer you are to more speculative investment opportunities- the more you’re also taught about “listening to your gut” and “developing conviction”.If you’re in public markets looking at stocks with years of recorded financial performance in an established industry, you’re looking at the same data points everyone else is looking at. Your job is to assess these data points, assign weights to each of them that reflect your opinion on the significance of each data point and the likelihood of any particular event playing out. In other words, you’re just guessing but with more steps. In private markets, your “gut feeling” plays a more prominent role, due to the scarcity of these data points. So you mold whatever figures you stumble upon into a narrative you’re comfortable with
All investment decisions have inherent biases baked in. But there’s trouble brewing when you don’t know what these biases are based on. You can dress them up in all the figures you want, get as granular as you want with your projections, but you can’t escape the bias.
Biases are grounded in all the materials you’ve consumed, the experiences you’ve had, the people you’ve spoken to. This is what shapes your individual, unique perspective and you should not run away from it, you should lean into it.
You know Databricks & Mortar is a challenger in an industry dominated by incumbents. But maybe this time you start by asking “what are incidents of challengers thriving in a competitive landscape?” Go on a treasure hunt and collect anecdotes.The index is not built for anecdotes
As an analyst in a financial services firm, your employer sets aside a chunky budget for all sorts of research software. You get access to industry reports, private data, you know… the works. They’re all great for grounding your perspective in data, but fall short at serving you anecdotal narratives. I previously wrote about building an index for financial analogies:
“If I want to research the consolidation of the streaming video market, I'll be asking several questions as starting points; has this happened before? In which markets and when? What were the drivers of that consolidation and does that match with what's happening in streaming video? Those questions will naturally translate to a series of Google search queries and a bunch of links. It's a lot of noise, a long filtration process to get to anything meaningful. I'll be left to my own accord to draw these connections. I didn't like my prospects, so I thought I should build my own library, create an index for my catalog that carries a deep understanding of the sources, and can draw relationships and connect the dots. In computer science, that's called a knowledge base.”
You’re always more likely to be wrong
You realize the error of your ways with Databricks & Mortar and choose to approach future prospective investments from a place of curiosity. You question your biases and spend time looking for anecdotes. You develop a narrative grounded in case studies, cautionary tales and you’re no longer using data as a crutch. You make the call “That stock will double within a year” you convince yourself. Within a few months the company underperforms, the stock suffers and they get acquired for pennies on the dollar. You still got it wrong, and that’s okay. You realize that we’re all just guessing but with extra steps. At least your steps are different from everyone else’s.