Models utilizing the Credmark Model Framework (CMF) before version 0.8.110 were confined to run within the context of a single network or chain ID provided in the input. With the introduction of context forking, a single model execution can now seamlessly execute code across multiple chains.


Every model in CMF requires a context for execution. When an end-user initiates a model run, no pre-existing context is available. So, a new context is created, primarily identified by the network (ethereum, bsc, etc) and block number provided in the input. Subsequent web3 calls, ledger queries, and other model executions within the originating model run within this created context. 

Previously, altering the context's block number or chain was not possible. However, the introduction of context forking addressed this limitation. A forked context is derived from the model's existing context. While the network of the new context is modified, the block number remains synchronized with that of the model's original context. Specifically, the block number of the forked context corresponds to the block mined just before or at the timestamp of the model's context. 

To illustrate, consider a model aimed at retrieving token prices from a centralized exchange (CEX). Without forking, the model would throw an error if the price was unavailable for the input chain. By leveraging the forking capability, however, prices can be sought across multiple chains, synchronized to the input block number of the original chain.

This enhanced functionality enables a more robust and versatile use of models within CMF, providing seamless interaction across various blockchain networks.

 runs a financial modeling platform powered by reliable 

 of quants, developers, and modelers actively build models for the DeFi community by leveraging our data API and tools. Join the growing community and together we will advance the next-generation financial system.


Interoperability Across Blockchain

Cross-Chain Models

Uniswap v3 introduced concentrated liquidity. This immediately caused two related problems:


It made it more difficult for liquidity providers (LPs) to deploy their capital.

It made computing pool performance metrics very difficult.


Impermanent Loss (IL) in an AMM pool is due to the unfavorable reduction in the amount of the more valuable asset when the pool changes its price. If the pool’s price is restored to the original value when LP deployed the position, the loss is eliminated, which is why the previous “loss” was called “impermanent”. 

Loss Versus Rebalancing (LVR)¹ has been proposed as a complementary performance metric that is computed by simulating a portfolio traded on a centralized exchange (CEX). This portfolio is kept identical to the LP’s pool position on a decentralized exchange (DEX) such as Uniswap v3. 

A detailed walk-through of this simulation is available in 

² Jupyter notebook. On the CEX we carry out the buy/sell actions necessary to maintain the asset ratio in the LP’s DEX positions. This is the act of rebalancing. We therefore have two identical portfolios (R for rebalancing portfolio and V for DEX portfolio) , in terms of asset composition, with different performances. This difference is the LVR at time t which can be expressed as:

We will show below how we implement this calculation for a Uniswap v3 pool by using its swap history.

The swap history for a Uniswap v3 pool fetched from its Swap events.

Using the Credmark Model Framework (CMF), we can


retrieve its Swap events for a range of blocks:

sqrtPriceX96: square of the ratio of two assets in X64.96 format that’s recorded by the pool³.

Although tick and sqrtPriceX96 can be converted interchangeably, sqrtPrixeX96 is more precise due to more significant digits. 

For the price after the swap, the price of token0, i.e. the amount of token1 for 1 unit of token0, is as follows:

The price of token1 is the inverse of the above:


The exchange rate for the swap can then be calculated as the following, as the ratio of the amounts in token0 and token1 excluding fee. We call them swap prices: swap0 and swap1. Swap0 is the amount of token1 for 1 unit of token0, Swap1 is the amount of token0 for 1 unit of token1.

The LVR is the difference in value of the DEX and CEX portfolios. 

Calculated taking into account the asset price after a swap and the swap price itself. This is an estimated optimal value calculated by computing the maximum difference between the price available on a CEX (we approximate it as the price after swap) and a price emergent on a DEX (the swap price itself). 

Note that we have negated the amount of the outgoing token to multiply with the price difference, meaning that we have performed the same operation from the pool to sell the token in both CEX and DEX and obtain the difference as LVR.

For a detailed walk-through, please refer to the Jupyter notebook⁴ mentioned earlier.

The above calculation has been added to our price dataset and we have completed a historical backfill for Uniswap v3 pools on Ethereum mainnet. A new model has been created to query this dataset and can be run using the Curl command below.

The output is a list of records. Each record contains the statistics for each interval of blocks.

By using this API, a user can compare the performance of DEX pools on Uniswap v3. 

To illustrate, we chose three pools for WETH and USDC/USDT/DAI respectively.

Using the data returned by the API, we calculate the ratio between the LVR and the earned fee. The smaller the ratio, the lower the LVR vs. the fee. The chart below illustrates this ratio for the three stablecoin pools.

USDC and USDT are about the same despite USDC having a spike in Sep-2023. The average for both of them is about 34%.

DAI had a spike in early 2023-03 due to SVB. This pool’s LVR is 4 times the fee because of slower price movement. If we exclude the spike, the average is still 47%.

Given the volume between the three pools, USDC vs USDT vs DAI is 6:2:1 and TVL at a similar ratio of 5:3:1, being an LP for USDC/USDT has similar profitability from trading fees and lower LVR than being an LP for DAI.

For a detailed walk-through, you may refer to 

In this post, we shared how we calculate the LVR for a Uniswap v3 pool and how we use it to compare the profitability of being an LP in different pools. We agree with the authors of the LVR paper that LVR should complement IL to track the performance of Uniswap v3 pools. 

We have made the historical and current data available through Credmark’s API. This complements the theoretical derivation in the paper. LPs can benefit from this dataset to further their investment analysis.

https://arxiv.org/abs/2208.06046?ref=blog.catalyst.exchange

https://github.com/credmark/credmark-models-notebook/blob/main/uniswap/lvr.ipynb

https://docs.uniswap.org/contracts/v3/reference/core/libraries/TickMath

https://github.com/credmark/credmark-models-notebook/blob/main/uniswap/lvr-bench.ipynb

 around the globe and across industries.

Our 

 of quants, developers, and modelers actively build models for the DeFi community by leveraging our data API and tools. Join the growing community and together we will advance the next-generation financial system.

Credmark's Dataset

Adding Uniswap V3 Pool Fees and LVR to Credmark’s Dataset

I’m happy to announce that, from today forward, Credmark will no longer be using the word “models”. Instead we will use “transformations”

, models and transformations are synonymous.

At Credmark we have always taken the most inclusive possible definition of a model. Our models could be simple functions, e.g., f(x) = x + 1, or as complex as a function that returns the historical performance of a portfolio.

This range of model types was confusing but manageable, until everyone started talking about AI models as if they were the only type of model.

AI models are models that are created by processing training data through a framework. When these models are run over new data, a large collection of operations are performed on this data to produce an output. These operations weren’t directly coded by humans but, rather, derived from the training data.

AI models are therefore a very specific type of model. Unfortunately, explaining that f(x) = x + 1 and a recent ChatGPT model are in the same class is difficult, and likely to be misunderstood.

This is why we will only use transformation from now on.

Transformations are models, but we are now going to leave the term “models” to AI practitioners.

“Transformations” is a better term for the data analytics field.

We don’t intend to rewrite history. Blog posts that use the word model will not be updated. Everything else – documentation, code repository name, code, websites – will be updated, slowly.

This will be a little confusing, but we can’t update all these materials overnight. Please bear with us as we work through this transition.

The Change

From Models to Transformations

Factory contracts are a pattern that DEX protocols use to efficiently manage functionality, governance, and extensibility. They automate the deployment of liquidity pools, enabling quick and efficient pool creation. They also provide transparency into a platform’s liquidity by serving as a comprehensive ledger for deployed pools. For developers building in DeFi, understanding factory contracts is key to fully leveraging the capabilities of on-chain liquidity.

How Factory Contracts Streamline Pool Deployment

At the core of DEXes is the factory contract. These smart contracts allow for easy deployment of liquidity pools. The process is simple: invoke the factory contract with the right parameters to deploy a pool contract. 

By doing so, users can freely create a pool for any token pair while the protocol controls the mechanics. 

This functionality enables token creators to quickly create  pools tailored to their tokens. For DeFi data analysts, factory contracts offer insights into the protocol's deployed pools.

DEX protocols use factories differently. Uniswap and QuickSwap have one factory per protocol version. Curve takes a different approach. For each pool version, Curve has two deployment methods. 

The registry allows the Curve core team to deploy pools. 

The other factory contract is permissionless, allowing anyone to deploy pools. 

This balances security with decentralization.

Curve's Ethereum mainnet deployment has four factory contracts:

: a registry of custom StableSwap pool (Curve v1) implementations deployed by Curve Core.

: a permissionless StableSwap pool (Curve v1) factory.

: a registry of custom CryptoSwap pool (Curve v2) implementations deployed by Curve Core.

: a permissionless CryptoSwap pool (Curve v2) factory.

Curve deployments on other chains may have different configurations.


To get a full list of Curve pools, we have to aggregate pools found in all four factories and eliminate duplicates¹. This provides comprehensive coverage while eliminating redundancy between Curve's controlled registries and permissionless factories. 

Having the list of all pools in Curve paved the way for us to inquire about one account’s LP positions in Curve.

DEX factory contracts enable frictionless liquidity pool creation. Curve’s blended approach, which supports both centralized and decentralized pool creation, addresses conflicting ecosystem needs and enables developers to build faster. 

As DeFi continues to grow, it is important to understand foundational design decisions that empower DeFi users and developers.

¹ We aren’t sure why duplicates occur, but they do! Our assumption is that sometimes a permissionless pool is recognized by the core team.

An overview into Curve factory contracts

How Curve Factory Contracts Streamline Liquidity Pool Deployment

In the analytics industry, the terms “data”, “datasets”, and “data series” are used interchangeably. So are “models” and “transformation”.


The distinctions between these are both subtle and not terribly important. But we should understand why.

 are a collection of discrete pieces of information. This information may be correct, or it may not be. Data are used as input for reasoning. Good, or correct data is likely to yield accurate conclusions; incorrect data is not. Of course, we all know that the same data can be used to reach several, sometimes contradictory, conclusions. Confusing. 

 are collections of data that are likely to be related. For example, all of the final grades of the students in classroom B.


 are datasets that represent a sequential pattern. For example, the price of a stock over a period of time is a data series.

From an analytics perspective, input data are input data. How related they are is just one of the data’s characteristics.

 are encoded actions that transform data into conclusions which, as stated earlier, may or may not be accurate. A conclusion may be a new data set. 

 are encoded actions that transform data into a new data set. The new data set may be a conclusion. 

The terms “model” and “transformation” are completely interchangeable. Data engineers tend to use “model” because their thinking is data-based whereas data analysts and data scientists tend to use “transformation” because their thinking is process-based. 

Since Credmark has built a framework for managing processes, we’ll use transformations to be clear.

Any kind of data manipulation is a transformation. There are 4 fundamental types: 

While trying to build up a consistent time series, we might use interpolation to 

While preparing a dataset for analysis, we may 

When processing a set of US addresses, we may 

 street addresses and zip codes into separate datasets for better analysis. 

To produce a dataset of unnatural deaths, we may 

These are, of course, primitive operations. Advanced transformations are likely to combine two or more of these basic transformation types. 

All of the following are transformations: 

Analysis is the application of transformations over datasets with the goal of reaching conclusions.

It uses data to find patterns or answer questions, which may allow for better decision-making. For instance, by analyzing sales data, we might discover that sales increase during certain seasons, leading to better stock management.


 sit at the heart of the analysis process. Transformations allow data to be prepared and interpreted. 

Put simply: analysis = data + transformations + intelligence. 

DeFi data is complicated, which makes analysis difficult. Credmark builds powerful DeFi-aware transformation tools to support analysis.

Data, datasets, data series, models, and transformations

What is Data Transformation?

Artificial intelligence (AI) is no longer a distant dream or a concept confined to the realms of science fiction. It's a reality that's transforming every facet of our lives and businesses. One area where AI is making significant strides is in the world of finance, specifically in the burgeoning field of cryptocurrencies. The crypto market, with its volatility and complexity, presents a unique challenge and an equally unique opportunity for AI.

The beauty of the crypto world lies in its transparency and the vast amount of data available on-chain. Every transaction, every movement of funds, every interaction with a smart contract is recorded on the blockchain, providing a treasure trove of data for AI to learn from. However, the sheer volume of this data can be overwhelming. It's like trying to find a needle in a haystack. You need to know what you're looking for, and you need the right tools to sift through the noise and find the valuable insights hidden in the data and be able to train AI and machine learning (ML) models.

Training an AI model is not a straightforward task; it's a complex process that requires a careful balance of various elements. Here are some of the key challenges and necessities:

The foundation of any AI model is the data it's trained on. The quality and quantity of this data can significantly impact the model's performance. In the crypto world, there's no shortage of data, thanks to the transparency of blockchain technology. However, the challenge lies in filtering out the noise and identifying the relevant data points. This is where platforms like Credmark come into play, offering curated data sets that can help train more accurate models.

Understanding the meaning of data can be important when training a model. We need to be able to distinguish between a mint and a burn event, for example. Naively decoded smart contract data doesn’t provide these labels. When possible, Credmark always provides this critical context.

In the context of AI, features are individual measurable properties or characteristics of the phenomena being observed. Selecting the right features is crucial for training an effective model. In the crypto space, these features could be transaction amounts, wallet addresses, timestamps, or smart contract interactions. The challenge is identifying which features are most predictive of the outcomes you're interested in.

Overfitting occurs when a model is too complex and starts to learn the noise in the data, resulting in poor performance on unseen data. Underfitting, on the other hand, happens when the model is too simple to capture the underlying structure of the data. Striking the right balance is crucial for creating a model that generalizes well to new data.

While complex models like deep neural networks can provide high accuracy, they often act as "black boxes," making it hard to understand how they're making predictions. This lack of interpretability can be a significant issue, especially in a field like finance where transparency and understanding are crucial.

The crypto market is highly volatile and can change rapidly. An AI model trained on past data might not perform well if the market dynamics change significantly. Therefore, it's essential to continuously update and retrain the models to keep up with the changing market.

Navigating the complex landscape of AI model training in the crypto space can be daunting. The challenges are numerous, from ensuring the quality and relevance of data to selecting the right features, avoiding overfitting and underfitting, ensuring model interpretability, adapting to market dynamics, and complying with regulatory and ethical standards. However, these challenges are not insurmountable, especially with the right tools and resources at your disposal.

Enter Credmark: Providing Crypto Data for AI

Credmark offers a comprehensive solution for harnessing the power of crypto data to train AI models. Their platform allows developers to create DeFi (Decentralized Finance) risk models faster than ever before. The integration of curated data sets, Python, and a library of pre-built data structures and models provide an incredible springboard for AI development.

Credmark's platform enables developers to focus on product iteration instead of building their own web3 data stack or creating custom integrations to wrangle data. This means more time can be spent on refining the AI models and less on dealing with the intricacies of data management.

This API provides access to granular token data made for deep analysis and decentralized applications (dApps). It includes current and historical token prices, decentralized exchange (DEX) liquidity, holdings, and token metadata. This data can be used to train machine learning models to predict token prices, understand liquidity patterns, and more.

This API allows you to trace all wallet activity and manage your portfolio with integrated risk models. It provides token price and balance, wallet activity, and risk metrics. This can be particularly useful for building AI models that predict portfolio risk or track wallet activity patterns.

This API is a gateway to DeFi, purpose-built for quants, modelers, and web3 builders. It provides access to over 200 financial models with one endpoint and allows you to create your own models with web3's most robust modeling platform. This API can be used to build sophisticated AI models that leverage financial data from the DeFi space.

Credmark also provides raw blockchain data as well as derived data, which can be used to monitor or trace any activity from the genesis block. This data can be used to train AI models on a variety of tasks, such as detecting fraudulent transactions or understanding blockchain activity patterns.

Credmark, with its comprehensive suite of tools and services, is uniquely positioned to help both seasoned and budding AI practitioners overcome these challenges and unlock the full potential of AI in the crypto space. One of the challenges with other data integrations is the difficulty in accessing and processing blockchain data. Credmark simplifies this process by providing decoded, indexed data that can be easily accessed from a Python environment. This means that data scientists and developers can use familiar Python tooling to work with the data, which can significantly speed up the model development process.

Moreover, Credmark's platform integrates with all the necessary data science libraries and supports Jupyter Notebook, making it an excellent tool for developing and training machine learning models. The platform also allows developers to easily write their own models and query historical data sets via APIs using an open-source framework, which can save time and resources that would otherwise be spent on building a custom data stack or creating custom integrations.

In conclusion, the data that Credmark offers is a powerful tool for training AI models in the crypto world. By providing a rich, curated, and accessible source of crypto data, Credmark can help to drive the next wave of AI innovation in the crypto space. The future of AI and crypto is bright, and platforms like Credmark are at the forefront of this exciting intersection of technologies.

A closer look into Credmark's Data Offering

Harnessing the Power of Crypto Data for AI

Web3 aims to distribute control and enable open access. Yet the crypto industry today relies heavily on pricing from centralized entities. Such concentration of power undeniably undermines the same values Web3 promotes.

Centralized exchanges (CEXes) can manipulate token prices, trading volumes, and momentum. In 

 blog post, I showed how Binance was caught red-handed when they tried to “instill confidence” during the UST crash.

Here’s why decentralized pricing is the solution.

Entirely derived from on-chain DEX data, it is:


Verifiable - Anyone can validate the pricing

Resistant to manipulation - Pricing comes from decentralized trades

Without decentralized pricing, Web3 cannot fulfill its promise. The same problems plaguing traditional finance will persist.

Without decentralized pricing, Web3 is still largely dependent on legacy systems and fails to fully break away from centralized intermediaries. The prices of crypto assets today are predominantly derived from centralized exchanges like Coinbase and Binance or data aggregators like CoinMarketCap and CoinGecko. This allows exchanges to manipulate markets for their or their customers’ benefit.

We’ve seen clear examples of such manipulation during events like the UST and the LUNA collapse. Centralized exchanges like Binance reported artificial price floors for UST while its peg was failing and paused LUNA price feeds as its price crashed. These misleading prices led to huge losses for traders relying on them.

From a Web3 perspective, it makes no sense to depend on centralized exchange pricing. Web3 is designed to create a more decentralized and transparent environment. Centralized exchanges are inherently centralized and opaque.

Arguably worse, centralized crypto exchanges lack the same oversight applied to traditional finance exchanges. Without regulation by authorities like the SEC, centralized crypto exchanges face little accountability for their actions and can manipulate the markets with impunity.

Smart contracts need prices, so they depend on Oracles (primarily Chainlink) because computing those prices from chain data would incur unacceptable gas fees. Chainlink and other oracles synthesize multiple prices provided by centralized exchanges. A source of prices purely derived from chain data is needed. Luckily 

Credmark uses on-chain DEX data to produce transparent, verifiable pricing for tokens. These decentralized prices are derived from liquidity and swap data from Uniswap and other DEXes. The pricing algorithm and model are completely open source so users who rely on them can verify the dependencies. 

Decentralized pricing will unlock Web3's true potential to transform finance and the internet as we know it. Without it, Web3 innovation will remain centralized and therefore vulnerable to manipulation by centralized players. 

Credmark built its price database and API to solve this problem. As Web3 and DeFi continue to evolve, decentralized pricing will be the key driver in ensuring the ecosystem lives up to its democratic ethos.

 of quants, developers, and modelers actively build models for the DeFi community by leveraging our data API and tools. Join the growing community and together we will advance the next-generation financial system. 

The push for decentralized pricing is crucial

Web3 Won't Work Without Decentralized Pricing

A new form of scam has emerged in recent months, tricking users into using 

scammers’ addresses when sending tokens. The most recent incident occurred on June 7, 2023, in block 17426333 on the Ethereum network.

In this article, we hope to shed some light on how this happens and provide insights to help other web3 participants so they can protect themselves against such attacks.

The attacker employs a systematic approach to carry out this scam. First, they scan recent ERC-20 token transfers, specifically looking for large amounts of ETH, USDT, or USDC transfers. Once the attacker identifies a compatible transaction, they create a fraudulent transfer that closely resembles the preceding transaction. This fake transfer appears legitimate in the owner’s wallet. The transfer address used appears to be identical to the address used by the legitimate transaction, although it isn’t. If a user copies the last-used recipient address without double-checking to send a token, the scammer will receive it instead of the intended receipient.

The fraudulent transfer mimics the original transfer in two ways. 

An attacker-owned recipient address, which is designed to resemble the same address from the sender, with the same starting and ending. That’s what most wallets show for the shortened version of the address. Even EtherSscan showed only the starting of the address till an update around Q1 2023. The attacker makes use of this common UI design to make the owner believe that the legitimate and illegitimate addresses are the same.

A fake token that mirrors the name of the same token used in the initial transaction.

As these scams become more common, it is crucial to verify the accuracy of the recipient's address before proceeding with any transfer.

The attacker likely used an address calculator tool, which allows them to quickly compute similar addresses with matching starting and ending sequences, along with their corresponding private keys. 

To gain a deeper understanding of the attack, one on-chain record of the incident can be seen here:

https://etherscan.io/address/0x80de381c38f96a97813a08a2c841168d472bf65f#tokentxns

We’ve gathered information regarding some fake token addresses associated with this attack. These addresses include: 

: 0x2e60f55e05320f16a06198eb7e27ad31ab57cf5e

: 0x599826f808af6c44cc65820c413aa32210246e21 

We’ve also obtained details regarding the individual behind these attacks, known as the "phisher," and their associated smart contract:

: 0x7B321d35F7D8732DEA643Bc2d727BE53c9912050

: 0x38e74D71Bce2e3eb50B22D74dC2D6E308e657d6D

As DeFi progresses onwards, new threats will continue to emerge. This recent copycat phishing scam  serves as a reminder of the importance of being cautious when making transactions. To safeguard your funds, it is crucial to verify recipient addresses as scammers will continue to find ways to game the system. By staying informed and following best practices, users can more safely navigate the industry.

A new form of wallet scam

Copycat Scam

In the previous two articles, we have laid out the two central designs principles of the pricing model:

This article presents a fix to the pricing model made necessary when the majority of the stablecoins in our Ring 1 depegged on March 10th 2023. On that date both USDC and DAI temporarily lost their peg which broke previous assumptions of the model and caused us to incorrectly price crypto assets.

To counter high inflation, the US Fed has been quickly hiking interest rates on a streak since March 2022.. Prior, significant capital found its way to DeFi where higher yields were available. As rates rose, a capital flight from DeFi to more traditional interest-earning assets. It didn’t happen. Ironically, DeFi was hurt following catastrophic failures in centralized crypto (e.g., FTX) and traditional finance (e.g., Silicon Vallye Bank). The fall of SVB was a monumental event, the biggest bank failure since the 2008 banking crisis.

On Friday March 10th financial regulators closed Silicon Valley Bankand took control of its deposits.

Circle, the issuer of USDC, a very popular stablecoin, announced that it had $3.3B of assets with SVB. Over the weekend, USDC lost its peg to the USD and fell as low as 86 cents. Another major stablecoin, DAI, which held a significant amount of USDC as collateral, also lost its peg.

On Sunday, the U.S. Treasury, Federal Reserve and Federal Deposit Insurance Corporation announced that depositors of SVB would have access to all of their money starting Monday.

After the announcement, Circle said it would have “full access” to its reserve in SVB and USDC would remain redeemable 1-to-1 with the US dollar. USDC began to regain its peg on Monday. DAI followed.

This depegging event impacted our pricing model fundamentally. The selection of the ring 1 stablecoins has assumed that there would be enough liquidity between the tokens in our ring of tokens to offset the depegging of the minority of the stablecoins so that we can value the fiat value of the stablecoins. Unfortunately, liquidity-weighted USDC and DAI made up more than ⅔ of our pool and they lost their peg.

When both USDC/DAI depegged, our base assumption became invalid and our stablecoin pricing failed. As we show in the tables below, at the start of depeg period at block 16802605 on Ethereum, Credmark assumed that USDC and DAI had held their peg resulting that USDT was overpriced. Our derived prices drifted by 5% to the Chainlink oracle’s price because of this error. As the distance from the peg became more severe at later block 16804605, our prices drifted by as much as 11%.

Since stablecoin prices are at the base of our pricing model, all other token prices drifted as well.

At block 16802605 (Mar-11-2023 04:55:23 AM +UTC)


At block 16804605 (Mar-11-2023 11:40:35 AM +UTC +UTC)

Since we set out to derive our prices entirely from DEX data, we necessarily depend on the relative ratio between stablecoins to derive their value. This works perfectly in a depeg scenario as long as the majority of the stablecoins hold their peg. This wasn’t the case, so our assumption was incorrect. We needed to rethink our assumption.

We have come up with another way of valuing our selected stablecoins in Ring 1. We introduced the notion of a cap. With the cap in place we would have adjusted the value of USDC and DAI and would have therefore priced stablecoin to fiat correctly.

In practice, stablecoins can fail and fall to USD 0 (see UST), but they can never be worth significantly more than USD 1. At times stables can be worth slightly more than USD in DeFi market. They have been due to short-term sentiment of buying, lack of liquidity or lack of arbitraging activities, but we can ignore such small fluctuations.

This insight allows us to correctly value stablecoins from their relative values a

s long as at least one stablecoin in our Ring 1 holds its peg

 since it allows us to rescale the other token accordingly.

Between any two stablecoins, we determine their relative value from the liquidity-weighted ratio in all pools of them.


We put the pair-wise ratio into a matrix.

Because we have three sets of ratio data, there is more data than we need. To find one set of ratios to be used, for each column, we calculate max/min range between the values and choose the one set with the smallest range to avoid data noise.

For the chosen set, we set the maximum value as $1 and rescale the rest to obtain the values for the stablecoins.

For the following pair-wise relative value matrix, the smallest range is the 2nd column of 1.05254226.

We set the maximum value in that column as $1 and rescale the rest according to the ratio.

Recalculating our examples shown earlier, the results are greatly improved from the previous model. The difference between our model and Chainlink oracles has been greatly reduced to the acceptable range. The residual difference lies in the difference between CEX and DEX data.

At block 16804605 (Mar-11-2023 11:40:35 AM +UTC +UTC)

During the pegged period, the previous model performed slightly better than the updated model but the difference of +/-0.3% is acceptable.

At block 16862071 (Mar-19-2023 01:29:59 PM +UTC), one week after the SVB event

The implementation has been included in our models repository 

https://github.com/credmark/credmark-models-py

 to calculate the prices for stablecoins. Such as, 

https://github.com/credmark/credmark-models-py/blob/main/models/credmark/protocols/dexes/uniswap/uniswap_v2_meta.py

Credmark Blog

For the latest technical analysis, partnerships, and community updates