Coinbase Deep-dive and Investment Thesis: Part 3

Coinbase valuation and investment thesis.

Disclaimer: The information and views expressed in this article are those of the author only and intended for educational purposes. They do not constitute any form of advice or recommendation to be relied upon in making (or refraining from making) investment decisions.

Key takeaways

  • Based on the results of my valuation model (and reserving some allowance for modelling uncertainty), Coinbase shares appear to be priced at or below fair value as of the writing of this article.

  • The current share price of Coinbase reflects only the value of its existing transaction and custody / staking businesses, and disregards the potential revenues from the expansion of its institutional and subscription-based services.

  • If I were to believe in Coinbase’s ability to successfully scale its recurring revenue businesses, there could be significant upside to my valuation.


In the first and second parts of this deep-dive series, I looked at Coinbase’s value proposition and went under the hood to dissect the operational and financial aspects of its business. In this third and final part, I tie everything together to build an investment thesis for the company.

The valuation model I built is hosted on Github and made freely available for download and (non-commercial) use. I i invite you to go through my work, test out different assumptions, and overlay it with your own analysis. Valuation is a necessarily subjective exercise—so if you think that your version improves upon mine, please do share your ideas and findings in the comments below. I’d love to hear them.

Motivation

Coinbase’s direct listing has proliferated wide speculation over the ‘correct’ value of its shares. A cursory review of headlines from major news outlets suggests that Coinbase may be overvalued by media consensus - but what evidence exists to support their assertions?

  • Why you can expect Coinbase to have a very hard time justifying its $61 billion market cap” – Fortune, 15 April 2021

  • Coinbase is Not Worth $100 Billion But Its Stock Could Double On Direct Listing” – Forbes, 13 April 2021

  • Opinion: Should you buy Coinbase? The valuation is ridiculous, based on this math” – MarketWatch, 12 April 20211

As a financial analyst by training, I was naturally curious to put my understanding of the company to the test.

In the remainder of this write-up, I discuss my valuation approach and step through my modelling assumptions. I then explain the results of my valuation (highlighting a few important caveats), before finally setting out my investment thesis for Coinbase.

Valuation approach

Analysts normally rely on one of two approaches when valuing a company:

  1. the comparables approach, which looks at the value of similar, publicly listed companies and scales this by some fundamental measure of size (for example, financial earnings or the number of accounts). The resulting multiple is then applied to the subject company to derive an estimate of its value. The comparables approach should only be used where the subject and comparable companies share similar growth and risk characteristics.

  2. the discounted cash flow (DCF) approach, which models the expected future cash flows of a business and discounts them back to the present date, is a more explicit valuation method. The discount rate used in a DCF reflects the time value of money and non-diversifiable risks that a company faces.

Coinbase is a market leader in an emerging sector. While it is comparable to other businesses in the crypto asset exchange space (e.g. Binance, Kraken), none of these companies are publicly listed and so I am unable to derive comparable multiples using them.

Some analysts compare Coinbase to other non-crypto asset exchanges or financial service companies. I believe this is an apples-to-oranges comparison and should not be done. As I explain in Part 1 and Part 2 of this series, Coinbase is exposed to materially different risk characteristics and revenue drivers to a conventional company. Therefore, I do not rely on any valuation metrics based on this group of companies either. Instead, I explicitly model Coinbase’s cash flows using a DCF.

Modelling assumptions

One of the first conscious decisions I made when building the valuation model for Coinbase was to avoid expressing a view on the price of Bitcoin —I thought that, tethering my valuation to such a speculative and dubious assumption would surely undermine the credibility of my analysis from the outset.

Yet, as I started to write Part 2 of this series, it dawned on me that Coinbase’s revenues are inextricably tied to the price performance of the underlying crypto assets (at least over the short- to medium-term, and possibly beyond). This meant that any robust forecast of revenues would need to be based, at least in part, on a forecast of Bitcoin’s price. This presented the first major modelling challenge in my analysis.

A second challenge came in the form of modelling the growth of retail and institutional transactions. In Part 2 of this series, I identified that institutional trading volumes have (historically) been less reactive to price volatility. I would expect this trend to continue over time, and so I needed to allow for greater volatility in retail transaction revenues without explicitly working out the price volatility of Bitcoin. This, too, informed certain design choices with respect to the modelling of revenues.

A third and final challenge came in the form of estimating revenues from Coinbase’s early-stage services, which do not currently generate a meaningful level of sales on which I can base my forecast. Brian Armstrong has stated that subscription-based revenues could eventually make up around half of Coinbase’s revenues, meaning the revenues from these services could be very material if Coinbase is able to execute on its vision. As I explain below, I ultimately do not forecast these revenues because there is simply no reliable way for me to do so. This is a key limitation to my analysis, and affects how my valuation results should be interpreted. I discuss this in more detail toward the end of this article.

Taking into account the above issues, my key modelling assumptions with respect to forecasting Coinbase’s revenues are as follows. I assume that:

  1. the price of Bitcoin will grow in line with the Stock-to-Flow (S2F) model (a popular asset pricing model in the crypto space), adjusted for the extent of over-/under-valuation and allowing for diminishing volatility over time (more on this later). 

  2. Coinbase’s Monthy Transacting Users (MTUs) will grow in line with the price of Bitcoin, as per my forecast. This growth reflects the onboarding of new users to the cryptoeconomy, as well as increased levels of engagement for existing users, over the forecast period.

  3. Coinbase’s average revenue per transacting user (ARPTU) will remain broadly constant at its historical 2019-2020 level. To the extent that Coinbase is able to continue to scale its revenue capture per user, this is likely to be a conservative assumption.

  4. a fee compression factor will kick in from Q4 2023 (shortly before the next halving of Q1 2024) as activity in the cryptocurrency market picks up again. Starting with a fee compression factor of 100% (no reduction) in Q3 2023, I reduce this factor by 5 percentage points per quarter until it reaches 35% in Q4 2026. At this point, the margin between Coinbase’s retail and institutional users is mostly eroded.

  5. Coinbase’s institutional transaction revenues will grow in line with the unadjusted S2F model (i.e. without applying an adjustment for over-/under-valuation). This is because I assume that institutions invest across the market cycle and represent a more stable source of revenues for Coinbase.

  6. Coinbase’s custodial fee, staking revenue, Earn revenue and interest income will grow in line with Assets on Platform, which in turn will grow with the price of Bitcoin.

  7. other revenue (comprising sales of Coinbase’s own crypto assets and interest income) will remain broadly at the same level, growing only by inflation. In effect, I assume that the amount of cash needed to run Coinbase’s business will not change substantially over time.

  8. other subscription and services revenue are non-recurring (i.e. fall to zero). As explained above, I am unable to estimate future revenues from Coinbase’s Analytics, Commerce and other subscription-based services because Coinbase does not currently report revenues from these segments, and I have found no reliable indicators on which to tether my forecast. To the extent that Coinbase is able to scale these businesses in the future, there is likely significant upside to my valuation.

To forecast costs, I assume that:

  1. transaction expenses will be 12.5% of net revenues. This is based on management expectation that these expenses “will be in the low-to-mid teens as a percent of revenue in 2021”.

  2. sales and marketing costs will be 14% of net revenues. This is higher than historical levels and is based on the mid-point of the 12-15% range provided by management.

  3. technology and development costs and general and administrative expenses will total $1.5 billion in 2021, based on the mid-point of the range quoted by management. I assume that these costs will then grow at 5% per annum, to allow Coinbase to continue to scale its business (particularly its institutional offering).

  4. other operating expenses (comprising costs from the sale of Coinbase’s own crypto assets) will remain broadly at the current level, growing only by inflation.

  5. pre-tax profits will be taxed at a rate of 21% (the US corporate tax rate)

  6. depreciation and amortisation will be roughly offset by capex in each period. This is reasonable because most of Coinbase’s capital investments are in technology assets that have a relatively short duration.

  7. working capital needs will grow in line with revenues as the company builds up a net receivables balance with its customers. To calculate working capital as of end-2020, I assume half of the cash and equivalents held by Coinbase at that date was surplus to its operations.

I model Coinbase’s revenues and costs over an explicit forecast period up to 2026, and calculate a terminal value to reflect cash flows thereafter. To do so, I assume a terminal growth rate of 2% (in line with inflation). Given the cyclical nature of Coinbase’s business, it is important that the terminal period reflects cash flows generated in a representative year over the market cycle, rather than a peak or trough. 2026 lies between two halving dates and satisfies this condition, based on my analysis.

Before presenting my results, I discuss two further aspects of my valuation in detail: how I forecast the price of Bitcoin, and how I estimate a discount rate for Coinbase.

Bitcoin price forecast

Bitcoin as a unique asset in that its monetary policy (in terms of the distribution of new supply) is pre-programmed into its code. Approximately every four years, a ‘halving’ event occurs in which the block reward distributed to miners (i.e. the rate of supply) is cut in half, increasing the scarcity of the asset. The table below summarises past halving events and the dates at which future halvings are expected to occur, assuming the duration in days between halvings remains constant.

One of the best-known asset pricing models in the crypto industry is the Stock-to-Flow (S2F) model developed by PlanB, an anonymous former institutional investor based in the Netherlands. The S2F model is based on the theory that the scarcity of an asset can be quantified by the ratio of its stock (in terms of the amount of available supply) to its flow (the annual rate of production). The S2F model has been successfully applied to other scarce commodities like gold and silver, and has done a surprisingly good job of modelling Bitcoin’s price.

The figure below shows that, on a logarithmic scale, the stock-to-flow value of Bitcoin is linearly correlated with its market capitalisation.

A simple linear regression of the logarithmic value of Bitcoin’s market cap (y) to the logarithmic value of its S2F value (x) provides the following model:

(For statisticians, the model is robust with an R-squared value of 0.916.)

Using the model above, and assuming a block is mined every nine minutes (the historical average rate), it is possible to calculate the expected future S2F value of Bitcoin and construct a price forecast for Bitcoin.2 The forecast results are shown below.

A couple of observations:

  • over the long term, the model has done a relatively good job of predicting the price of Bitcoin

  • after each halving event (as shown by a vertical increase in the S2F value), there has always been a subsequent but more delayed increase in the price of Bitcoin

  • the grey bars show the extent of over- and under-valuation relative to the S2F model. If we believe the S2F model to be fundamentally sound (i.e. reflecting the fair value of Bitcoin), it can be seen that Bitcoin often enters periods of excessive exuberance whereby asset prices rise significantly above the level predicted by the model.

  • the extent of over- and under-valuation is not constant, however. As shown by the peaks and troughs of the grey bars, the extent of over and under-valuation (i.e. volatility) has been reducing over time.

In my experience, the peaks and troughs of Bitcoin’s price, however interpreted, are important characteristics of the market cycle that cannot be ignored. The S2F model does not capture these peaks and troughs. Therefore, in my price forecast I have applied an additional adjustment for the extent of over- and under-valuation, assuming past cycles repeat but allowing for diminishing volatility over time. In particular, I assume the extent of over-/under-valuation follows the level observed 46 months (i.e. one halving) prior, but reduce this level by 10% for each subsequent halving.

My adjusted S2F forecast is shown in the figure below, superimposed on the S2F price forecast.

Discount rate

The discount rate for a company is often calculated as a weighted average of the return on its equity and debt. Coinbase does not hold any debt on its balance sheet, so I am only required to estimate its cost of equity.

A common framework used to estimate the cost of equity is the capital asset pricing model (CAPM). It looks at the sensitivity (or beta) of a company to the broader market to estimate a discount rate based on non-diversifiable risk.

Coinbase has only traded for around a month as of the writing of this article. That is not enough data to derive a robust estimate of beta. While it is normally possible to use the returns of a portfolio of Coinbase’s listed comparable companies to proxy Coinbase’s returns, as explained above, it does not have any suitable comparators. Therefore, I use Bitcoin’s price as a proxy for Coinbase’s.

The 5-year monthly returns from April 2016 to March 2021 imply a beta for Coinbase of 1.30. This is in line with my expectation that the company is more exposed to market volatility than the average company.

At the current 10-year treasury yield of around 1.5%, and assuming an equity risk premium of 5.5%, I estimate a cost of equity for Coinbase of 8.7%.

Valuation results

Bringing everything together, the table below summarises my forecast of Coinbase’s revenues and costs.

After discounting these cash flows back to present value, my model values Coinbase at around $81 billion at the current date. This is equivalent to a fully diluted share price of $405, or around 35% higher than the current share price of $298 as of the writing of this article.

There are a number of caveats to this figure:

  • the results of my model are highly sensitive to the modelling assumptions I have used, many of which are subjective. For example, if I were to reduce the extent of over-/under-valuation in my adjusted S2F price forecast by 20% (rather than 10%) for each subsequent halving, my valuation would fall to $74 billion or $372 per share. If I were to increase my discount rate to 10%, my valuation would fall to $65 billion or $325 per share.

  • my model assumes that, from 2027 onwards, the price of Bitcoin will remain fixed at $375,000, growing only by inflation. If Bitcoin’s price continues to rise over each subsequent halving, my model is likely to understate the value of Coinbase.

  • my model crucially omits revenues from Coinbase’s subscription-based products, because I am unable to build a reliable forecast of these revenues. To the extent that Coinbase is able to scale its revenues from these income streams, there is likely substantial upside to the value I have estimated.

In light of all my findings, I summarise my investment thesis for Coinbase in three points:

Investment thesis

  1. Based on the results of my valuation model (and reserving some allowance for modelling uncertainty), Coinbase shares appear to be priced at or below fair value as of the writing of this article.

  2. The current share price of Coinbase reflects only the value of its existing transaction and custody / staking businesses, and disregards the potential revenues from its institutional and subscription-based services.

  3. If I were to believe in Coinbase’s ability to successfully scale its recurring revenue businesses, there could be significant upside to my valuation.

That brings us to the end of this three-part deep-dive on Coinbase. If you enjoyed the article, please consider subscribing to get notified whenever I publish a new deep-dive.

Til next time!


Subscribe to the blog

Hi, I’m Park!

Open Source Finance is a blog to document my journey into the complex and fast-moving world of fintech. In this blog, I share ideas, insights and key learnings about companies, technology, entrepreneurship, innovation and more.

Finance is better when it’s open source. If you enjoy Open Source Finance, please consider subscribing and sharing it with your social network to help me build my profile!

1

The article uses something called a ‘reverse DCF model’ which backs out the implied revenue growth rate and profit margin based on assumptions about return on invested capital (ROIC) and cost of capital (WACC). While the finance theory behind this is sound, it is difficult to assess what assumptions are embedded in this analysis and whether it is sufficiently robust.

2

When calculating the circulating supply of Bitcoin, I exclude the first million coins (mined from January to September 2009) to account for lost coins. This is the same adjustment originally applied by PlanB.