Models
Disclaimer: This is not financial advice. Anything stated in this article is for informational purposes only, and should not be relied upon as a basis for investment decisions. Chris Keshian may maintain positions in any of the assets or projects discussed on this website.
To subscribe to this mailing list, input your email here.
One hurdle investors face when allocating to crypto is finding appropriate valuation models. This is further compounded by the often frustrating behavioral dynamics that affect prices in this space. For example, any investor who had a well-reasoned investment thesis, complete with accurate projections of future cash flows and a comprehensive DCF model, was likely outperformed by Dogecoin last cycle. And while Doge is a silly, unique case, the dynamics that underpin its success highlight some of the nuances of investing in an asset class that is liquid, early stage, and largely community-owned.
Below I will walk through a brief history of a few crypto valuation models, and end with a linear regression model that I use as a starting point to form my understanding of price action and value accrual in this space.
In 2017, John Pfeffer wrote a seminal piece on crypto valuation models called “An Institutional Investor’s Take on Crypto Assets”. He proposed valuing network tokens (e.g. Ethereum) like a nation state currency.
M = PQ/V
M is the total money supply; that is, the total number of coins
V is the "velocity of money": the number of times that an average coin changes hands every day
P is the price
Q is the transaction volume: the economic value of transactions per day
The value of a single token = M/T, where T is the total number of tokens.
This is a good starting point, but as the space has matured and different token models have surfaced, this valuation problem has become more complex. In addition, the liquid nature and retail ownership of tokens in this asset classes introduces behavioral economic principles that complicate the above mathematic purity.
In reality, there are numerous variables that affect token price. I have highlighted some of these complicating factors below:
Liquid Markets - Tokens are liquid and actively traded. These are early stage VC investments that you can trade 24/7.
Massive Retail Market - Anyone with an internet connection can on-board and begin speculating.
No-arbitrage - Many assets do not have the balancing market effect of short interest, as there are not perp or futures contracts for most tokens.
Hype Cycles - These assets are subject to wild hype cycles, often driven by hodl zealots on Twitter and Discord, and get bid far higher than any reasonable fair value.
Different Fundamental Valuation Models - Even though people outside of the space tend to lump all cryptocurrencies into one bucket, in reality there are a wide range of token use cases and value accrual mechanisms.
Reflexivity - This asset class is highly reflexivity, both on the way up and on the way down. This further amplifies variance - creating much more severe boom bust cycles.
Power Laws - Globally accessible networks are unique and the power law applies. The value of these networks can increase exponentially once flywheels kick in, so the TAM used to value these businesses further compounds speculative frenzy.
There have been many attempts to develop valuation models for crypto, which I classify in three main buckets: absolute valuation models, relative valuation models, and traditional valuation models.
Absolute Valuation Models - Valuing crypto relative to other asset classes or scare resources
Value crypto as a percent of the remittance market it can capture.
Value crypto as a percent of the Gold market it can capture.
Value crypto as the amount of attention it can capture. Dogecoin captured the attention of the world. Attention is a scarce resource, so anything that captures attention has value.
Relative Valuation Models - Valuing crypto relative to itself
Plan B’s Stock-to-Flow Model for Bitcoin.
Ethereum Flow-based Model - Similar to the above.
Supply and Demand Models, like the framework Ray Dalio used in his long-form piece on Bitcoin.
The Rainbow Model (more on this below) - Which superimposes a rainbow on the Bitcoin log chart . This model divides the bitcoin price into eight colored bands: bubble, sell, FOMO, bubble formation, HODL, still cheap, accumulate, buy and discounts.
Hodl Waves - An on-chain valuation method that assesses holding behavior and the velocity of coins in the bitcoin economy. It shows the time periods in which bitcoin has most recently moved.
Traditional Models - Applying valuation models from equities or commodities to crypto
DCF - Tracking protocol revenue, projecting future cash flows, and building a model to determine the present value of the project.
Valuing certain projects based on "buy and burn” models that function similar to stock buy backs in public equities.
Value crypto as a commodity
The Bitcoin Rainbow Chart
Models as Memes
While there are hard metrics that are useful in valuing many projects, some of the other, more “hand-wavy”, valuation models listed above make trading and investing in this asset class particularly challenging.
Models like the Bitcoin Rainbow Chart (pictured above), can be packaged into memes - accessible, easily digestible units with high viral potential - and propagated across the internet to massive audiences. If one of these models is easy to understand, and has high viral-potential, it can quickly attract buyers. When enough market participants who are struggling to value these assets buy into these “model memes”, they can have a self-fulfilling prophecy. You can see how, if enough Bitcoin investors bought into the above Rainbow chart, the maximum number of market participants would buy Bitcoin in the blue zone and sell it as it approaches the red zones, causing price to follow this model in each successive cycle. Again, this may seem silly, but collective belief is a powerful force and should not be written off lightly, even if the object of that belief seems juvenile.
These models then become Schelling Points for market participants, as they collectively attempt to price certain crypto assets. But the problem with these Schelling points is that they are predicated on a collective belief, and sometimes that belief breaks down.
Stock-to-flow was effectively broken earlier this year. Knowing this and understanding the price point at which believers in this particular valuation model would capitulate was a useful edge.
The narrative that Bitcoin was an inflation hedge broke down when the CPI print hit 6.2% last November, crypto spiked and then plummeted. It has been in a downtrend since, while CPI has been rising, temporarily destroying the “inflation hedge” meme.
As demonstrated by these examples, part of the challenge is not just thinking about what the valuation model is, but also knowing what models other people are using to value certain assets. Specifically as a trader, knowing where the market is overestimating or underestimating certain factors can generate alpha. This is often referred to as the “metagame” or the game within the game, which can be described as subset of the game’s basic strategy and rules that are required to play any game at a high level. Understanding what metagame is being played, and what the incentive structure is behind the game, can give participants an edge.
The Power of Narrative
In the absence of rigorous quantitative models, narratives become increasingly important. In his book “Narrative Economics” Robert Shiller discusses the virality of narratives and their profound impact on the economy. No where is this narrative force more palpable than in crypto, where the rapid dissemination of information amongst the millions of speculators across Twitter, Discord, and Reddit can ignite and propagate a narrative that gains mass adoption and unified belief in a matter of days.
A few examples of narrative fervor:
•2020 DeFi Summer - The narrative was that DeFi tokens can be valued using the NPV of the value that accrues to the token via fees. TVL, it was said, could be a proxy for platform value. But the overlooked question was - is this TVL just mercenary capital?
•2021 Alternative L1s - The narrative of the “Eth-killers”. Ethereum is slow and cost prohibitive for new users. These new smart-contract platforms are better. SOL returned 12,000%, so AVAX will too, and FTM, and ….
The capital rotation game that ensued as these narratives played out was remarkable, and front running these rotations proved wildly lucrative.
While the above examples have longer narrative arcs, some short-term narrative trades are event driven. Just last week, in the depths of a crypto bear market, Doge increased more than 150% on the news that Elon Musk had successfully taken Twitter private. The narrative around this was that Musk would find some way to integrate the coin into the platform.
My Approach
Each of the previously-mentioned models are imperfect. They are like the story of the blind men attempting to describe an elephant based on which part of its body they were touching. But a comprehensive model that accounts for all of these elements could drive us closer to the truth.
To understand these mechanics and to begin formulating my own investment thesis, I created a linear regression model that exhaustively captures all aspects of the above models, while also incorporating the nuances of this asset class and the wealth of data sources available. The goal of this exercise was not to create a new valuation model for crypto assets, but rather to determine which of the independent variables listed above had the greatest impact on price. This approach helped me delineate the variables I should pay attention to, and weight more heavily in my analysis, from those that can be ignored.
I began by asking a high level question - If I could create a master formula for all of the variables that affect the price of a given crypto asset, what would the formula look like? I did this exercise for Layer 1 tokens, but the same thought process can be modified and applied to other crypto verticals.
I started with the below, reductive formula:
Price = Fundamental Value + Capital Inflows + On-Chain Metrics + Demand - Supply + Technical Analysis + Sentiment Analysis + Noise
I then broke these high-level buckets down into component parts, and constructed the below linear regression model. This model is specific to Layer 1 tokens, but the framework can be modified with different metrics for other verticals. This model regresses the change in price (Δprice) against all of the independent variables that might act on price.
Linear Regression Model
ΔPrice =
(fundamental value)
+ total number of unique developer accounts making commitments to chain-affiliated repos
+ code complexity of GitHub commits made to affiliated repos
+ collective value of all dapps on chain
+ user numbers for dapps on chain (DAU, MAU)
+ collective volume of transactions across all dapps on the Layer 1 chain
+ net inflation/deflation
+ TPS or some measure of time it takes to transact
+ % toward carrying capacity, i.e. saturation of chain (developers/transactions)
(on-chain metrics and capital inflows)
+ long-term holder accumulation
+ number of unique wallet addresses on the chain (metcalfe’s law)
+ number of active wallet addresses on the chain
+ size of developer incentive fund
+ Net Unrealized Profit/Loss of token holders
+ collective value of stablecoins (dry powder) on chain in a given ecosystem
+ stablecoin flows into or out of a given ecosystem
(demand)
+ variable quantifying how many people currently want to buy this asset
+ variable quantifying purchasing power of individuals/entities who want to buy this asset
+ has there been a reasonable change in value for this project (more market share, so project makes more on fees, etc.)
+ variable quantifying a meme army - coalition of activists/promotors/buyers in the community
(supply)
+ hodl waves - is there a reason people want to hold longer (staking, rewards, etc.), driving a decrease in supply
+ inc/dec in token economics (changes like EIP-1559)
+ token unlock schedule
+ no arbitrage - e.g. people cannot short the asset because there is no perp contract
(technical analysis)
+ open interest
+ funding
+ RSI
+ Stoch RSI
+ EMAs (12, 21)
+ MAs
+ S/R levels
+ MACD
+ AO
(sentiment analysis)
+ variable quantifying narrative potency
+ meme virality
+ meme stickiness
+ sentiment score for Discord community
+ google ngrams
+ Twitter propagation
(macro)
+ interest rates
+ QE/QT
+ M2 - crypto is highly correlated with money supply
+ Risk Assets strength
+ 10 yr real rates
+ 10yr-2yr rates
Not all of the above independent variables were easy to quantify, but I made a best effort. I trained this model on subset of data from 2018 - 2022, and then tested it on a different subset of the same time series. I then ran this model with different subsets of the independent variables listed above. I also ran this model multiple times, using different data and time frames to distill a list of the most significant independent variables.
I then filtered out the variables with the highest significance scores to determine which had the greatest impact on the change in token price.
I repeated this process for seven different Layer 1 tokens, and cross referenced the most significant variables for each instance to generate a final list. The result was a collection of independent variables that I use to value crypto assets, which I will elaborate on in future posts.