This is the second post in a series about stock valuation methodology. Previously I shared how to reconcile deep skepticism of approaches like Stockfisher with how I ended up using it.
Last time, I wrote about how Stockfisher uses the value investing approach made famous by Warren Buffett and his mentor Benjamin Graham. It estimates the intrinsic value of a company.
In this piece, I will explain how we estimate instrinsic value. I start by defining what intrinsic value is. I then explain the forecasts that we use to estimate it. I'll then state the assumptions we use, and ultimately how we arrive at the Stockfisher fair value for every company in the S&P 500.
What is intrinsic value?
The intrinsic value of a company is defined as the discounted value of all the cash that a business will generate for its owners during its existence.
Having estimated the intrinsic value for a variety of businesses, a long-term investor can then buy stock in those businesses that are trading at a stock price that is less than the intrinsic value.
A key advantage of this approach is that it allows a long-term, independent minded investor to ignore the market. If an investor is able to accurately forecast the future cash produced by a business and has a long-term horizon, then they are free to ignore the gyrations in its stock price. Instead, they can sit back and simply collect the cash produced by the business, exactly as if they held ownership in a private business!
Of course, a key difficulty all lies in understanding a business well enough to make accurate long-term forecasts! It is easier said than done.
[An important aside: When we say accurate forecasts, we mean accurate probabilistic forecasts. Like a weather forecast with a 90% chance of rain. For a continuous variable, we assign 80/20 confidence intervals meaning that if the forecast is well-calibrated then the outcome will land inside the interval 80% of the time.]
While many investors agree with the intrinsic value philosophy of investing, the future is quite uncertain. It is time-consuming and error-prone to project out financials for decades. So, effectively no one does this. There are a variety of common short-cuts that investors, including myself, use.
How do investors currently approach valuation?
It is easy to find consensus revenue & earnings estimates out to 1-2 years. Large compnies often attract 20+ analyst forecasts of their quarterly earnings. Interestingly, most of those same analysts would agree that estimating the 5 to 10 year revenue and earnings is far more important for determining the fundamental valuation. It is just too hard, and there is strong demand from the market for short-term projections and analysis.
Among the long term projections that I’ve seen, they often end with something like: "We assume a 35x exit multiple in 4 years" or a relative comparison to industry peers. And, there is no shame in this, I do this myself! However, it also is pretty far from estimating intrinsic value.
Finally, it is easy to find great spreadsheet models of a business’s economics over time from great analysts. While these models are highly uncertain, they can help illuminate the most sensitive assumptions to figure out the crux of an investment. Often different industries are modeled using different metrics and key indicators. I am a huge believer in this approach and I often personally make simple models like this.
With Stockfisher, our primary goal is to efficiently value every company at the highest possible quality. This would allow us to harness the power of AI to value every company in a neutral way that allows for a systematic comparison and sorting of all stocks based on up to date information.
What Stockfisher is trying to accomplish
We don’t make any claim that our valuation methodology is superior to a financial model that is based on KPIs and unit economics for a business. Quite the contrary. We agree those specialized models offer superior analysis and insights.
When we started our goal was to make a model that is:
- Generalizes to all companies
- Results in an actual cash flow model over the life of a business, to actually estimate its intrinsic value
- Has scorable probabilistic forecasts, so that calibration and accuracy can be objectively measured.
- Is approximately right, even if we know it isn’t perfect
- Has high quality long-term forecasts, based on deep research
This would provide a consistent, neutral methodology, with transparent rationales to support the forecasts and valuation, all focused on the long-term. It would be largely immune from sentiment and short-term considerations. So it would give reasonable valuations that can be sorted and constantly updated based on news and changing prices.
I think we’ve achieved the above with Stockfisher!
The rest of this piece explains the details behind how Stockfisher comes up with its valuations.
The key stock valuation parameters: revenue, net margin, and payout ratios (dividends + repurchases)
For Stockfisher, we wanted to value all companies using a common framework. That means that it can’t rely on industry specific metrics.
We experimented and considered a few decompositions, but ultimately landed on forecasting revenue, net margin, and shareholder payout ratio. We like this breakdown because each of the three components has different characteristics and considerations that go into forecasting it.
If you multiply Revenue x Net Margin (Earnings/Revenue) x Shareholder Payout (Cash to Shareholders/Earnings), the result is cash paid to Shareholders. Said another way, if you know the Revenue, Net Margin, and Shareholder Payout ratio for all time, then you can calculate the cash paid to owners for all time! So, we have what we need to calculate intrinsic value - assuming the forecasts are accurate. And of course, this allows us to do what we wanted - sort by discount to intrinsic value given today’s stock price.
If we have a cash flow projection over the life of a company, then we can alternatively calculate the IRR (internal rate of return) that an investor would earn if they bought at today’s price and held it indefinitely! I like to call this the intrinsic return.
OK, so how do we get forecasts of revenue, net margin, and payouts forever? We have to make some simplifications and assumptions.
5 and 10 year forecasts
Forecasting the long-term is hard. But, it is also essential if you believe in intrinsic value.
Rather than simply substitute an easier question (what will next quarter’s earnings be?) for the hard question, we attempt to tackle the hard question head-on. This doesn’t mean that we’re under any illusion that we can predict exactly what revenue will be in 10 years. Quite the contrary, we simply attempt to assign the narrowest possible confidence intervals that are still well-calibrated. And often those intervals are quite wide.
We make probabilistic forecasts of the future, just like a meteorologist might assign a 90% chance of rain in 3 days. A key benefit of making probabilistic forecasts is that they are precise and can be scored for accuracy and calibration. I’ve spent over a decade of my life doing judgmental forecasting as a Good Judgment Project Superforecaster and at Metaculus. Future posts will go more into how we draw from this literature and practice.
For Stockfisher, we had to decide what time scales to forecast and how many forecasts to make.
One quarter and even one year into the future don’t weigh too heavily in any calculation of intrinsic value. Nor do we think we can do better than Wall Street already does.
Ideally, we would love to forecast 20+ years into the future. But, at that time scale things are so uncertain as to become too hard - since your forecast might need to consider the possibilities of radical technological or geopolitical change.
We settled on forecasting revenue, net margin, and payout in both 2030 and 2035, roughly 5 and 10 years from now. This felt like a good balance between being important for intrinsic value and being able to forecast something meaningful that isn’t too distinct from the world we know today.
Note that unlike most Wall Street projections, our 5 and 10 year forecasts are resolvable and can be used to measure our calibration. And, if there were other consensus probabilistic forecasts, we’d be able to objectively compare the accuracy of our forecasts to the consensus. But, we know of no such consensus forecasts. (Please let us know if we’ve missed them!)
To estimate the intrinsic value, we need projections for every year, not just 5 and 10 years out. We need to make some assumptions.
Interpolating fundamentals over the next 10 years
From now until 2030, we simply interpolate.
For example, if a company’s revenue is $100 in 2025 and forecast to be $200 in 2030, we use exponential growth of 15% per year to fill in the revenue. In 2026, the projected revenue would be $115. This seems like an incredibly reasonable approximation to us for the vast majority of companies.
For net margin, we interpolate linearly between the net margin today and the forecast value in 2030. If the net margin is 10% in 2025 and projected to be 15% in 2030, then we use 11% as the net margin in 2026. Again, this feels like a great approximation to me.
For payout, we interpolate linearly like net margin, but we have a few additional guardrails to cover some edge cases. For instance, we assume the payout from 2026 to 2029 can never exceed the forecast payout in 2030. This addresses some cases where the 2025 payout might be 300% - clearly an unsustainable amount.
From 2030 to 2035, we interpolate between the 2030 and 2035 forecasts. Again, this feels quite uncontroversial to me.
At this point, we have what we think is a pretty reasonable model for the payments to shareholders over the next 10 years. For many companies this is a substantial portion of where their intrinsic value comes from.
While forecasting out to 10 years is far further than Wall Street consensus extends, we still need to project out further to complete our calculation.
To do that we need to make some terminal assumptions.
Assumptions about the long-term future
A key goal of ours was to try to make a model that is neutral between high-growth companies where the intrinsic value might mostly come from cash generated more than 10 years in the future VS stagnant companies that are spewing cash today, but have little prospect for growth. Both types of companies can have high intrinsic returns depending on the price paid.
To try to fairly balance this, we assumed that revenue growth from 2035 to 2040 is half the rate of revenue growth between 2030 and 2035. This gives the companies with rapid growth more time to grow in our model.
While it is easy to find analyses that make high terminal growth assumptions, this makes me uncomfortable. In 2021 Warren Buffett noted that none of the top 20 biggest companies by market cap in 1989 were still in the top 20!
To attempt to balance competing considerations, Stockfisher’s model assumes 0% revenue growth starting in 2040, that net margins remain stable at their 2035 value, and that the payout ratio is 100% starting in 2040. It isn’t perfect, but we think it approximates the natural progression as high-growth companies mature and start returning more capital to shareholders. At the very least, this feels more principled to me than simply assuming a 35x exit multiple in 4 years.
At this point, we now have achieved our goal above!
We have a generalized model, a consistent methodology, and projections of cash flows over the life of a company.
Putting it together to calculate intrinsic value
Given the projected cash flows and the market cap today, we have everything we need to calculate the IRR that a long-term shareholder would earn if the projections are accurate!
To calculate the intrinsic value, we need one final ingredient: a discount rate.
We don’t think Stockfisher is informative of whether the market is over- or under-valued. So, we choose a discount rate of 6%. This makes the market approximately fairly valued. Said another way, there is roughly a 50/50 split between over and under valued stocks according to Stockfisher using a 6% discount rate. And, we’ll plan to adjust this rate as time goes on to maintain this balance.
Transparency in the model
Stockfisher shows the trailing 10 year financials as well as the forecast values in the Forecast tab, so you can see exactly what is happening. Here are the forecasts for Apple, as of today:
The supporting research behind these forecasts can be found in the News, Docs, and Management tabs. The rationales that support the revenue forecasts are provided in the revenue tab. Here is the 2030 revenue forecast rationale for Apple:
In the valuation tab, we show how the revenue, net margin, and shareholder payout % are combined for each year to project shareholder payout in $ over the life of the company. We also discount those payout $ with the discount rate, so you can see how the valuation is calculated by adding up the discounted payout $ for all time. Here is what this looks like for Apple:
Concluding thoughts
I think we have made something revolutionary for value investors.
We have found a way to estimate intrinsic value the way we are taught in Finance 101. We actually project the shareholder payouts over the life of every company using a transparent, generalizable methodology. This makes Stockfisher different from any other product.
Of course, the value of this approach is entirely dependent on the quality of the forecasts! The next piece go into how we forecast, and what we can know about accuracy.