Sensitivity analysis sweeps one input of a token model across a plausible range while holding the rest at base case, and records how far the output moves. It ranks your assumptions by consequence, which tells you where diligence is worth spending and where it is not. Across the token models we build, a small number of parameters carry most of the outcome, and the demand-side ones carry more than the supply-side ones the team has usually spent its time arguing about.
The output worth having is not the ranking, it is the cliff. Most parameters move the result smoothly and one or two have a threshold where the model flips from solvent to not, and that threshold is a number you can write into the mechanism as a control.
Sweep one input, hold the rest, record the swing
The procedure is deliberately unglamorous. Take the base case model. Pick a parameter, say the share of an unlock that reaches the market within thirty days. Run the model at the low end of a defensible range, run it at the high end, and record the change in whatever output you care about, usually cumulative coverage or the month the treasury runs out. Repeat for every parameter, then sort by swing size.
Two things make the result honest or useless. The ranges have to be defensible rather than symmetric decoration around the number you hoped for, because a narrow range on a parameter will make it rank low. And the output has to be the one that matters. Ranking parameters by their effect on peak market capitalisation produces a different and much less useful list than ranking them by their effect on whether the reward budget is still funded in year three.
Demand growth moves the answer further than anything else
There is a structural reason the demand assumptions dominate, and it is not that modellers are careless. Cong, Li and Wang's token valuation model makes user adoption and token price jointly endogenous: adoption drives price and price feeds back into the incentive to join, so the two move together rather than one being an input to the other.1 A parameter sitting inside a feedback loop does not move the outcome proportionally, it compounds.
Peer-reviewed token-economy frameworks treat pricing, stability and governance the same way, as joint outcomes of the parameter set rather than as independent dials.2 The practical read for a founder is blunt. If a model's answer changes materially when monthly active demand growth moves by a couple of points, then the model is a demand forecast wearing a mechanism costume, and the mechanism debate is not where the risk is.
Look for the cliff, not the slope
A smooth response is reassuring and mostly uninteresting. What you are hunting is discontinuity: the staking participation rate below which emissions exceed sinks, the price at which a fee-funded buyback stops covering its own gas, the unlock share above which modelled liquidity depth cannot absorb the sell without a price impact that triggers the next problem. Those are thresholds, and a threshold is a specification.
Once found, it stops being an analysis output and becomes a mechanism. A design that reads the parameter and adapts, by gating emissions on a measured sink rate rather than on a calendar, converts the cliff into a controlled behaviour. A design that only knows the cliff exists has documentation. That conversion is the reason to run the sweep at all.
One at a time is a real limitation, and it hides the case that kills you
Holding everything else at base case is what makes the method readable, and it is also what makes it incomplete. Real failures are joint. Demand comes in weak, and because demand is weak the price is low, and because the price is low the unlock recipients sell more of their allocation, and because liquidity is thin that sale moves the price further. A one-at-a-time sweep never generates that sequence, because it changes one thing while pretending the others sat still.
So pair it with something that varies inputs together. A Monte Carlo run over correlated inputs will produce the joint case. At minimum, define two or three named combined scenarios by hand, put the top-ranked parameters at their bad end simultaneously, and check whether the design still holds. The sweep tells you what to worry about. It does not tell you what happens when all of it goes wrong at once.
Common questions
What does sensitivity analysis show in a tokenomics model?
It shows which assumptions the outcome actually depends on. By sweeping one input at a time across a defensible range and recording how far the result moves, it ranks parameters by consequence. In token models the demand-side inputs typically dominate, because adoption and price are jointly endogenous rather than independent, so a demand parameter compounds through the model instead of moving it proportionally.1
What is the difference between sensitivity analysis and stress testing?
Sensitivity analysis varies one input at a time to rank assumptions by how much they matter. Stress testing pushes several inputs to a bad state at once to see whether the design survives a specific adverse scenario. They are complements. The sweep tells you which parameters to put in the stress scenario, and the stress test catches the joint failures a one-at-a-time sweep structurally cannot produce.
Which token model parameters usually matter most?
The demand-side ones, in our experience: growth in genuine usage, the share of holders who stake rather than sell, and how much of each unlock reaches the market. Supply-side parameters like emission curve shape matter less than the time teams spend on them, because they are chosen rather than uncertain. Run the sweep on your own model rather than inheriting this list, since the ranking is design-specific.
See Tokenomics Audit for how this applies in practice.
Sources
- Tokenomics: Dynamic Adoption and Valuation (NBER Working Paper No. 27222)
Lin William Cong, Ye Li and Neng Wang, National Bureau of Economic Research, 2020
Token valuation model in which user adoption and token price are jointly endogenous, the structural reason demand parameters dominate the ranking in a sweep. - Decentralized Token Economy Theory (DeTEcT): token pricing, stability and governance for token economies
Frontiers in Blockchain, 2023
Peer-reviewed framework treating token pricing, stability and governance as jointly modelled, parameter-driven outcomes.
Last reviewed 2026-08
Know the terms but not sure how they apply to your project? That is what an engagement is for. We design, document, and stress-test the whole token economy inside the Tokenomics Data Room.
100+ projects advised. Complete tokenomics in 4 to 6 weeks.