Financial markets are built around the powerful assumption that the future can be estimated. That assumption has created an enormous industry of analysts, quants, risk managers and algorithms designed to turn uncertainty into numbers. ‘Value at Risk’ models estimate potential losses, and backtests test strategies against historical prices. Liquidation engines calculate when collateral should be sold while credit models estimate who will repay.
Some financial events cannot be reliably assigned probabilities because the underlying system keeps changing; Frank Knight made this distinction more than a century ago. In Risk, Uncertainty and Profit, published in 1921, Knight separated measurable risk from uncertainty that cannot be reduced to known probabilities.
These systems are young, reflexive and interconnected. Their historical datasets are short, their rules change rapidly, with their worst failures creating conditions that have never existed before. The result is a dangerous possibility: a model can be mathematically correct and still be wrong about the world.
Knight’s Risk Is Not Knight’s Uncertainty
Knight’s distinction is simple but profound; risk exists when possible outcomes can be identified, and probabilities can reasonably be estimated. Insurance companies can model mortality because they have enormous historical datasets, just as a bank can estimate default probabilities across millions of borrowers. Uncertainty differs because it arises when the future includes events for which reliable probabilities cannot be calculated.
This is the foundation of risk vs. uncertainty finance. A DeFi protocol might calculate a 15% probability of liquidation based on historical volatility, but what happens if a new exploit drains liquidity from the protocol? What happens if an oracle fails? Or when three major lenders simultaneously liquidate the same collateral? The model may contain an answer, but the problem is that the answer may describe a world that no longer exists.
Why DeFi Magnifies Model Failure
DeFi creates unusually difficult conditions for financial modeling because its systems are interconnected through programmable contracts. A token can simultaneously be collateral, governance power, liquidity, and speculative exposure, while the same asset can appear across several lending protocols. A price collapse can therefore trigger liquidations, creating selling pressure that pushes prices lower and triggering more liquidations.
Bank for International Settlements (BIS) research finds that wallet leverage across major DeFi lending platforms typically ranges from about 1.4 to 1.9 times equity, with the largest and most active users carrying higher leverage. Higher leverage also increases the share of debt approaching liquidation, and the problem becomes worse when models depend on prices that themselves depend on the same market.
Mango Markets demonstrated this in 2022 when Avraham Eisenberg manipulated the low-liquidity MNGO market, pushing its price dramatically higher and then using the inflated collateral value to borrow roughly $117 million from the protocol. Chainalysis found that MNGO trading volume was about 2,000% above its previous ten-day average during the episode.
The protocol had not necessarily “failed” to follow its rules, but the assumptions feeding those rules had been attacked. That is classic model failure in P2P markets and DeFi. The model assumes prices contain trustworthy information, while an adversary has an incentive to make those prices lie.
The Backtesting Trap
Historical data feels reassuring because numbers look objective. A backtest takes old market data and asks a simple question: if we had used this strategy in the past, how would it have performed? That can help you understand volatility, drawdowns, and how a strategy behaves under familiar market conditions.
The problem is that the future does not have to behave like the past, and a strategy tested on five years of Bitcoin prices has not experienced five different versions of the future. It has experienced one historical path. The next five years could contain events that never appeared in that dataset, such as a major smart-contract exploit, a stablecoin collapse, an oracle attack, a regulatory shock or the failure of several connected institutions at the same time.
This is one of the biggest limits of financial modeling in crypto, and historical data can tell us how an asset behaved before, but it cannot tell us whether the assumptions behind that behaviour will remain true.
DeFi makes this problem even harder because financial protocols are connected to one another. A lending platform may depend on an oracle for prices, use another token as collateral and interact with several other protocols for liquidity. If one part fails, the damage can spread quickly.
The numbers from 2022 show just how serious this problem can become, with Chainalysis estimating that $3.8 billion was stolen from cryptocurrency businesses that year, and DeFi protocols accounting for 82.1%, or roughly $3.1 billion, of the total. Cross-chain bridges alone accounted for 64% of the stolen DeFi value. These were not simply bad trading days that a volatility model could have anticipated.
Oracle manipulation provides an even clearer example because DeFi protocols use price oracles to determine the value of assets used for trading, borrowing and collateral. If an oracle reports an inflated price, a trader may suddenly appear to own far more valuable collateral than they really do.
Chainalysis identified 41 oracle manipulation attacks in 2022 that caused approximately $403.2 million in losses. The Mango Markets attack showed how this could happen in practice, and in October 2022, Avraham Eisenberg manipulated the price of the relatively illiquid MNGO token. The artificial price increase made his position appear to be worth hundreds of millions of dollars on paper, and he then used that inflated collateral to borrow roughly $117 million in crypto assets from the protocol.
Knightian uncertainty in DeFi can assign probabilities to events it understands, but it struggles with events that change the assumptions the model itself depends on. A historical backtest might tell a lending protocol how much collateral it usually needs during a normal 20% price decline, but it cannot easily tell the protocol what happens when an attacker manipulates the price feed, drains liquidity and triggers thousands of liquidations simultaneously. Historical data is therefore valuable, but it has a boundary.
FTX Shows What Models Cannot See
FTX promoted sophisticated risk controls and an automatic liquidation system, but according to the SEC, Alameda Research was secretly exempted from important parts of FTX’s risk engine and was effectively given a virtually unlimited credit line funded by customer assets. No conventional market-risk model could compensate for undisclosed institutional behaviour, and the issue was not simply excessive leverage. The information entering the risk system was incomplete.
The US Department of Justice later said Sam Bankman-Fried misappropriated billions of dollars in customer funds and used them to support Alameda investments and other activities. He was sentenced to 25 years in prison and ordered to forfeit more than $11 billion. FTX therefore demonstrates an important limit of quant finance: models cannot calculate information that institutions deliberately hide.
SUGGESTED READ: What Happened to FTX? The Anatomy of an Implosion
Leverage Turns Uncertainty Into Fragility
Leverage is particularly dangerous because it compresses the distance between an ordinary price movement and insolvency. Suppose a trader posts $100 of collateral to borrow $80. A moderate decline may be manageable, but if collateral falls rapidly, the protocol must liquidate the position.
Now imagine thousands of traders holding similar positions; liquidations become market sales, market sales push prices lower and lower prices trigger more liquidations.
The Financial Stability Board has warned that crypto intermediaries combine familiar vulnerabilities such as leverage, liquidity mismatches, operational weaknesses, and interconnectedness, and that some combinations of functions can amplify those vulnerabilities.
Humility as a Design Principle
The answer is not to abandon financial models because the models remain essential. The answer is to stop treating them as descriptions of reality itself. Good financial systems should assume that models can fail. That means conservative leverage, collateral haircuts, circuit breakers, liquidity buffers, oracle redundancy and stress tests built around scenarios that have never happened before.
It also means designing systems that can survive periods when probabilities become unreliable, and that is the deeper lesson of the Knightian uncertainty in DeFi. Risk management asks, “How likely is this event?” but uncertainty management asks a harder question: What if our assumptions about the possible events are wrong?
Crypto’s greatest financial innovations may ultimately come from answering that question, as the strongest systems will not be those that predict the future with perfect precision but those designed to remain solvent, adaptable and transparent when prediction fails, and that is the real meaning of financial humility. A model should inform a decision, never become the reason a system forgets that the future remains unknown.
Disclaimer: This article is intended solely for informational purposes and should not be considered trading or investment advice. Nothing herein should be construed as financial, legal, or tax advice. Trading or investing in cryptocurrencies carries a considerable risk of financial loss. Always conduct due diligence.
Enjoyed this? Bookmark DeFi Planet, explore related topics, and follow us on Twitter, LinkedIn, Facebook, Instagram, Threads, and CoinMarketCap Community for seamless access to high-quality industry insights.
Take control of your crypto portfolio with DEFI PLANET PRO, DeFi Planet’s suite of analytics tools.
