# A Modular Adaptive Risk Layer for All of DeFi Concordia delivers a scalable modular Adaptive Risk Layer that mitigates market risk and paves the way for mass adoption. ### The Limitations of Static Risk Models in DeFi Static risk models, currently prevalent in Decentralized Finance (DeFi), pose a significant and multifaceted threat to the ecosystem's financial and developmental stability. Akin to the models employed before the 1987 stock market crash, these static models are vulnerable to systemic failures. One of the most poignant examples of this fundamental design flaw was the USDC depeg event, where instead of dynamically increasing LTVs, the static models allowed users to borrow more USDC, which could have fallen further. These vulnerabilities sometimes translate to substantial economic losses for users, with the Curve Aave, Mango markets, and Solend attacks totaling more than a quarter of a billion dollars in losses. These exploits ultimately hinder overall stability and trust within the DeFi ecosystem and render it susceptible to systemic risk, failing to capitalize on the learnings that traditional finance ("TradFi") gleaned over the past three decades. Furthermore, the inability of static models to adapt to the dynamic nature of DeFi significantly hinders the creation of novel, composable DeFi primitives. This directly impacts the potential for a modular defense strategy in the future, ultimately stifling the growth and mass adoption of DeFi by limiting its functionality and versatility. **Static risk models are a barrier to innovation, hindering DeFi's full potential to revolutionize financial landscapes.** * **Billions of Dollars Lost:** Systemic failure threats from static risk models still linger in DeFi, exposing billions in user funds due to potential model shortcomings. * **Prevents Innovation & Adoption:** Current inferior models limit the creation of new, composable DeFi primitives, stifling growth and mass adoption by hindering the development of a truly versatile and user-friendly DeFi ecosystem. ### Concordia: A Hub for DeFi Innovation Fueled by Adaptive Risk Management Concordia serves as a catalyst for unlocking DeFi's true potential by addressing core challenges and facilitating mass adoption. Through its modular Adaptive Risk Layer, developers can create groundbreaking applications, leveraging dynamic, AI-powered risk management for composable innovation. This proactive approach enhances resilience against systemic risks, safeguarding users and fostering trust. As a result, Concordia empowers a new era of institutional adoption, forging a path towards a secure, scalable, and inclusive DeFi future. Its innovative framework cultivates a robust ecosystem, encouraging widespread participation and driving DeFi towards its full potential. Functioning as a central hub for the next generation of DeFi innovation, Concordia's Adaptive Risk Layer offers unparalleled flexibility and adaptability. This enables builders to construct leading-edge projects within the ecosystem, promoting collaboration and pushing the boundaries of DeFi possibilities. Concordia's risk model creates a fertile ground for incubating cutting-edge DeFi protocols such as Superposition ([Superposition OnePager (Public)](https://docs.google.com/document/u/4/d/1tIAjcik7ykpcCF0UbRrrKF3VcBQOjHedshKo8wchDnM/edit)), paving the way for a wave of groundbreaking advancements yet to come. Each new protocol built upon Concordia further strengthens the AI engine, fueling its continuous learning and improvement through a powerful flywheel effect. # Portfolios Portfolios are risk boundaries of assets & liabilities Every collateral and debt belongs to a specific Concordia **portfolio**. A portfolio is the basic risk perimeter for a set of positions. The portfolio is a *unit* that is appraised as a whole: * A portfolio can take on new debt only if its set of collaterals can support it. * All of a portfolio's collaterals are cross-collateralized. If a portfolio is forcefully closed-out due to its risk quantity, *any* of the collaterals within a portfolio may be seized in order to rebalance its debts. Concordia portfolios give users an ability to separate their various positions into isolated groups. A single Concordia profile can have arbitrarily many portfolios. For instance, a certain user may wish to separate out a portfolio of long-term, low risk holdings from a different one that they use for high-frequency daily trades. ### Portfolio structure A portfolio is simply a set of **collaterals** and **liabilities**. The difference in market value between these amounts the equity in a portfolio. The specific allowed types of collaterals and liabilities are determined by the [collateral basket](broken://pages/tl0zUvjRFjnIYkbNDNVE)[ ](broken://pages/tl0zUvjRFjnIYkbNDNVE)that the portfolio is attached to. At creation, every portfolio selects from available collateral baskets, and the chosen basket determines which collateral types the portfolio may use. Since loan brokers also select a specific basket, the mutual basket in common between portfolios and brokers determines which liabilities a portfolio can take on. {% hint style="info" %} **Example** There is a small collateral basket that includes: * USDC * BTC * ETH There are 2 brokers that employ this collateral basket: * DOGE * SOL A portfolio elects to assign itself to this basket. This decision means the portfolio is locked in to only depositing collaterals USDC, BTC, and ETH. It can only take out loans from brokers that also use this basket. At this moment, the portfolio can only take on debt in DOGE and SOL. {% endhint %} ### Portfolio risk As a portfolio contains the pot of collateral that secures its set of liabilities, each portfolio carries a quantity of risk. The amount of risk attributed to a portfolio is a tantamount to a minimum amount of equity the portfolio must have in order for it to continue with its open positions. If a portfolio's equity falls beneath the risk threshold, the portfolio may enter into forced close-out. Managing the health of portfolio by paying down debts or adding collateral is the owner's responsibility. If at any point the portfolio does not meet its equity requirements, the Concordia network may automatically begin closing out the portfolio's debts in exchange for seizing a portion of the portfolio's collateral. # Risk Framework Concordia’s risk framework applies to the capacity to borrow assets and use them as collateral. It is a system to secure loans with collaterals in such a way as to balance capital efficiency and risk proper collateralization. ## Principles of Concordia's risk engine Concordia’s risk engine is: 1. Holistic 2. Data driven 3. Dyanmic ### 1. Holistic Concordia’s risk model appraises the risk of a **portfolio** as a whole. Liabilities and collaterals are not appraised separately, but rather in the context of their specific composition. A portfolio is a set of collaterals and liabilities. The ultimate question of the risk engine is whether (and to what confidence) the set of collaterals can secure against the liabilities. If the risk is too great, then the portfolio owner will either need to reduce their debt, pledge additional collateral, or suffer forced close-out. {% hint style="warning" %} **Key difference** In on-chain finance (as elsewhere in crypto-oriented trading platforms), the conventional approach to risk assessment is to assign haircuts to collaterals independently of each other, and then appraise whether the scaled-down value of the pot of collateral is sufficient to back the sum of outstanding debt. In Concordia, the quantity of risk is attributed to the *entire portfolio*, not the individual collaterals. To take an example from digital assets, Bitcoin is a high risk collateral for borrowing USDC, since the value of Bitcoin is extremely volatile relative to the value of USDC. But Bitcoin is relatively less risky when used as collateral to borrow ETH. The reason Bitcoin is less risky in the context of ETH is that the value of these two assets are highly correlated: if Bitcoin depreciates, it is a safe bet that so will ETH. {% endhint %} The specific positions are measured against their historical performance in order to quantify the amount of risk inherent in that portfolio’s basket of collaterals’ likelihood to secure that portfolio’s debts. Consequently, changes in the portfolio’s set of positions will change the haircut it receives. Every portfolio composition has a deterministic relationship to its specific quantity of risk. This approach to risk can create portfolios that are more capital efficient than on competitor’s platforms. For example, suppose a DeFi lending protocol assigns a 20% haircut to USDC (a dollar-pegged digital asset). If a user were to pledge $1 of USDC, they could only borrow up to $0.80 of USDT (another dollar-pegged digital asset). Concordia’s risk model, on the other hand, would inspect the concrete risk of backing USDT debt with USDC collateral, and since these assets have very similar identical historical performance, the would could be significantly less than 20%, presenting a Concordia portfolio with much more borrow power. ### 2. Data-driven On-chain finance happens in real-time and in a transparent, easily accessible public ledger. Market data is flowing out in the open. Concordia’s risk model reacts to the real-time changes in market conditions. It tunes its parameters automatically & algorithmically. As volatility (or other indicators) shift over time, Concordia's risk engine re-calculates risk based upon the changing markets. ### 3. Dynamic In order to avoid toxic liquidation spiral, a dynamic approach to rebalancing (liquidation) is required. Concordia use both Dynamic Closing Factor and Dynamic Incentives to increase incentives for liquidators by increasing the amount of collateral available for liquidation in line with the riskiness of the distressed portfolios' Health Ratio, thereby further de-risking the protocol linearly with the risk it carries. ## # Market Risk Concordia's risk engine responds quickly and accurately to market conditions. In a world where cryptocurrency prices fluctuate wildly from hour to hour, having a real-time model that can keep up with these changes is crucial for lenders looking to manage their risks effectively. At a fundamental level, the model requires the user to deposit collateral that protects the portfolio from the risk of the fluctuations of the asset prices over a certain period of risk (e.g., 1 day) with a probability (e.g., 99%) by using the appropriate filtered historical simulation VAR (value-at-risk). Below, we break this model into several steps, highlighting exactly how the risk model calculates the real-time price risk that drives its liquidation mechanism. **Step 1 -> Calculate the historical returns:** A filtered historical simulation VAR (value-at-risk) formula achieves a dynamic model. The model looks at the unweighted historical returns over a long time window (i.e., 180 days). The volatility measure implied from those returns alone would be similar to conventional protocols, which cannot note or predict unconditional volatility (i.e., future volatility not influenced by historical risk variables). **Step 2 -> Calculate Dvols (devolatilised returns).** By removing the volatility from these historical returns, we arrive at devolatilised returns (Dvols). This is an essential step because Dvols provide the unconditional volatility (not influenced by real-time variables) data needed to re-apply real-time, conditional volatility. **Step 3 -> Revolatilising Returns (Rvol)** Revolatilising returns (Rvol) attempt to predict real-time or conditional volatility. We do this by choosing a shorter time window (e.g., one day) to calculate the most recent volatility and apply that measure to the Dvols. We then get Rvols, a dynamic conditional vol measure that can more accurately pick up real-time price risks in the market.
Protocols that use static historical volatility measures have inefficient Health Ratio ratios that liquidate positions too late, creating toxic liquidity (bad debt). Moreover, the protocol loses potential loan revenue when Health Ratio ratios are too high because they use static historical measures. In summary, the filtered historical simulation VAR model turns down Vol when markets are calm and turns up Vol as markets become more volatile, enabling a more capital-efficient real-time model.
# Rebalancing (Liquidation) **Basic Concept of Rebalancing** A portfolio may be forcefully rebalanced if it has insufficient equity. Rebalancing a portfolio involves taking over a portion of its debt and collateral, with the effect of increasing its health. A portfolio's health is a ratio of its total risk over its total collateral. To calculate total risk, the market value of its liabilities are summed and added to a quantity of risk. In this formulation, "risk" is just the required equity in a a portfolio -- that is, the value of collateral in excess of the value of liabilities. $$ HealthRatio = \frac {Liabilities + Risk} {Collateral} $$ For a portfolio to be considered "healthy", its total collateral value must be greater than its liabilities added to the portfolio specific amount of risk. $$ Collateral \gt Liabilties + Risk $$ **Dynamic Close Rebalancing** The Concordia liquidation mechanism is one of the most vital components ensuring the protocol's health. Creating a successful liquidation mechanism must consider several primary concepts: 1. Preventing toxic debt 2. Ensuring adequately capitalized and decentralized liquidators 3. Other Design Factors. \*Note that neither is mutually exclusive and centered around incentives. **Terms** * **Bounty:** Also referred to as Liquidation Fee or Liquidation Spread. The discount is offered on collateral to liquidations to repay parts of the loan. * **Close-Factor:** the maximum amount of debt that can be closed out and repaid in a single liquidation, eventually equal to 1 in the dynamic close outlined below. # Preventing Bad Debt One of the primary points of consideration is preventing toxic liquidity, defined as liquidations that lower the health factor of the portfolio instead of improving it. As highlighted by the Curve Attack on Aave, static incentives pushed the portfolio into a toxic liquidity spiral that, with each subsequent liquidation, deteriorated the health of the portfolio as opposed to improving it, creating lousy debt. The solutions laid out by the [Toxic Liquidation Spiral paper by Jakub Warmuz, Amit Chaudhary, and Daniele Pinna](https://arxiv.org/pdf/2212.07306.pdf#:~:text=In%20this%20paper%2C%20we%20argue,liquidation%20spiral%20on%20the%20platform.) are two-fold (note that this assumes a traditional DeFi liquidation order book queue where liquidations bid for discounts on unhealthy collateral to bring it back to healthy levels by repaying parts of the loan): 1. Dynamic incentives 2. Dynamic Closing factor **Dynamic Incentives:** Dynamic Incentives ensure that liqudations can’t: 1. Make the position more unhealthy 2. Make the position healthier than the initial margin (except for Health Ratio < 1) 3. Can’t be more than x% \* collateral discount {% hint style="info" %} x% is defined by asset risk classification (the riskier, the higher the discount) {% endhint %} The point of dynamic is incentives to liquidate the portfolio in a way that prevents toxic liquidity - recall from above, this is not bad debt but a liquidation that deteriorates the health factor of the portfolio, which then typically leads to bad debt - insolvency. This equation can be written as $$(i) < (1/HealthRatio) -1$$ Where (i) equals the collateral value of the loan amount repaid plus the premium of the liquidation incentive. As the Health Ratio approaches 1, the liquidation incentives diminish, which is by design, as there are still instances in which liquidation arb opportunities would exist to close this position, especially in volatile mean reverting markets. This equation can also be written as a non-negative number; see equation 7 in the [Toxic Liquidation Spiral paper](https://arxiv.org/pdf/2212.07306.pdf#:~:text=In%20this%20paper%2C%20we%20argue,liquidation%20spiral%20on%20the%20platform.). Additionally, the liquidation queue limits liquidators from making the position healthier than the initial margin. Combined with the above formula, this creates a band that liquidators must operate within, thus protecting the borrows from unethical liquidations. Lastly, there is a liquidation fee cap that does not allow fees to go beyond x% of collateral. This is set by asset risk classification, with riskier assets being assigned a higher % discount cap than risky assets.
**Dynamic Closing Factor:** Under this model, the amount of the user's unhealthy portfolio available for liquidation increases as it gets closer to a Health Ratio of 1, rising to a point at which the entire collateral is offered to liquidators. Therefore, the dynamic closing factor further incentivizes liquidators in times of stress to compensate for the diminishing dynamic incentives as the portfolio gets closer to bad debt (see equation 8 in the [Toxic Liquidation Spiral paper](https://arxiv.org/pdf/2212.07306.pdf#:~:text=In%20this%20paper%2C%20we%20argue,liquidation%20spiral%20on%20the%20platform.)). In Corcodia’s case, the dynamic close can be set at a higher threshold, thereby closing out the entire portfolio before it reaches a Health Ratio of 1. This threshold is called the dynamic close, similar to margin sales in traditional capital markets. Note this threshold closeout variable is set by governance. In summary, the use of both the Dynamic Closing Factor and Dynamic Incentives together allows Concordia to increase incentives for liquidators by increasing the amount of collateral available for liquidation in line with the riskiness of the distressed portfolios' Health Ratio, thereby further de-risking the protocol linearly with the risk it carries. # Composite Operations In order to perform arbitrary actions with borrowed funds, Concordia allows for composite operations. A user may perform any sequence of operations in a single **atomic transaction** so long as the user’s portfolio is healthy at the conclusion of the transaction. This loose constraint – that the portfolio is healthy at the conclusion of the transaction – allows for the portfolio to become unhealthy at any point during the transaction. #### An example of composite actions is performing a “margin trade.” In this situation, the user is able to perform a composite operation where they: 1. Withdraw the entirety of their portfolio 2. Trade the withdrawn capital for another asset 3. Deposit that different asset into their portfolio For example: suppose a user has $4 of ETH in their account, with $2 of equity. In this situation, the portfolio is 2x leveraged. They have access to twice as much value as their equity. In other words, half of their portfolio's value is a liability. The user wants to trade the ETH for USDT at 2x leverage (essentially shorting ETH). In one transaction: they **withdraw** all $4 of ETH, swap it for \~$4 of USDC, and **deposit** the \~$4 USDC into their portfolio. At the conclusion of this transaction, this portfolio has $2 of ETH liability and $4 of USDC deposited. This kind of leveraged trade would not be possible without the ability to compose operations in a single transaction, since the withdrawal of all $4 ETH makes the portfolio unhealthy – though only “temporarily” when immediately followed by a deposit of $4. #### The most common pattern for composite operations is to perform an action with borrowed funds. They generally will take the shape of: 1. Borrow an asset in order to lever up one’s holdings 2. Withdraw assets from a portfolio (making it “temporarily” under-collateralized) 3. Perform an operation – or several – on a 3rd-party protocol with these withdrawn assets 4. Deposit a new sum of assets into the portfolio, restoring its health from the withdrawal at Step 2. But any operation on Concordia can be composed with others in a single transaction. Application developers are free to create useful compositions to create convenient financial products for users. Programming a composite instruction has blockchain specific requirements. Please see the developer documentation for instructions on how to create multi-operation transactions. # Pricing Engine How does Concordia collect and stream real-time prices Concordia’s pricing system provides real-time 1-minute granularity information on digital assets. It is designed to provide accurate spot prices with low latency, detect anomalous behavior in price movements, and gracefully smooth out corrupt or missing data. ### Aggregation algorithm For every instrument tracked by the Concordia pricing engine, there are numerous high quality data providers that measure market data to produce a current spot price. Different instruments will have different sets of data providers – for example equities data will come from different sources than cryptocurrency data.
In order to narrow the confidence interval and find an accurate price out of the collection of different data providers, the pricing engine filters the outliers using a standard interquartile range (IQR) filter. This method is very robust against outliers resulting from errant or corrupt data providers. After filtering outliers, Concordia takes the median of the remaining set. The reason to choose the median rather than the mean is to make the final quote less sensitive to outliers. In the case of price data, the expectation is that statistical dispersion is more likely a sign of a set of corrupted (or latent) data providers, rather than true volatility in the markets. ### Price record elements Concordia’s pricing engine constructs aggregations and statistical measurements out of the raw 1-minute streaming data. These include: * Current spot price - The median of the confidence interval for the price * Last observed timestamp * 1h return (delta since 1 hour prior) * 24h return (delta since 24 hours prior) * 1 hour moving average. The average price for the last 60 minutes. * 1 day moving average. The average price for the last day. * 1 hour half-life exponential moving average of price * 1 day half-life exponential moving average of price * T-score on a 20 day t-distribution of hourly returns * T-score on a 90 day t-distribution of daily returns * 1h return volatility - rolling exponential moving average * 1d return volatility - rolling exponential moving average ### Historical data The history of all Concordia’s price data is stored at hourly granularity. Users can retrieve historical data to analyze trends on all data points. ### Application to quantitative risk The elements in the price datum were included with risk assessment in mind. For applications that need to put the price in context of its past, these several data points can help to paint that picture. Here are some ways the Concordia Risk Engine uses these data in practice: * **New position circuit breaker**. A market manipulator can take advantage of naive risk engines by moving the price of a collateral suddenly up, borrowing funds, and then letting the price of the collateral drop back down to a “true” market price. Even if the attacker defaults, and gets their portfolio liquidated, they may still profit from the borrowed funds they pocketed. There is a reverse attack where they push the price of an asset down, and borrow a large amount of it at an unnaturally reduced value. Concordia uses the t-score on returns to decide whether or not a price movement is bonafide, and then prevents new positions from being opened on abnormal changes. When a 1h return is above the 99% p-score for the instrument, Concordia will not allow anyone to add to their collateral balance in that instrument. When a 1h return is beneath a 1% p-score for the instrument, Concordia will not allow anyone to add to their liability balance in that instrument. Once the hourly changes converge back within a nominal range on the t-distribution, positions can be modified as usual. * **Portfolio valuation clamps**. The value of collaterals and liabilities are “clamped” within ranges of the exponential moving average of the instrument. For instance, if the reported price of Bitcoin is $44,000 while the 1h EMA is $40,000, then Bitcoin is 10% above its EMA. The risk engine may clamp Bitcoin’s price to 5% within the 1h EMA, treating Bitcoin as effectively worth $42,000 for purposes of portfolio valuation. Not only does this protect against price manipulation, but it helps mitigate the effect of flash crashes or flash pumps on borrowers who would need time to make margin calls. Without price smoothing, inappropriate liquidations could occur when a price suddenly “wicks” in one direction or another. ### Price Sources * Pyth: * CoinGecko: * Switchboard: ## API Documentation The API for the price warehouse is hosted at # Diversification Risk The web3 space has a high level of operational risk that arises from assets built on top of protocols that might harbor contract bugs, exploits, rug pulls, and team and regulatory risks. These risks are typically not captured in historical data and can create a spurious correlation between assets. The Correlation add-on allows the protocol to limit the margin benefit to multi-asset portfolios that over-rely on historical data. An example of this would be a portfolio consisting of two stable coins. The long-window historical volatility shows close to perfect correlation; however, we subjectively know that one protocol contract might be flawed or the regulatory framework surrounding its pegging mechanism could be at risk. Therefore, the protocol can make a subjective decision to negate any correlation between these pairs. Given this is DeFi, Concordia will take a more conservative approach and limit most of the correlation benefits in multi-asset portfolios.
# Concentration Risk Concentration risk is another facet that needs to be added to DeFi risk models. Traditional exchanges look at the liquidity of instruments in a portfolio to limit exposure to heavily concentrated or illiquid assets. The idea of prominent concentrated positions adding more risk is relatively simple. Bigger positions and less liquid assets can cause price impact, making it harder to liquidate safely, which can cause bad debt. Concentration Charge works similarly to those seen in the trad-fi markets: It's a charge added to the filtered historical simulation VAR price risk calculation to account for the price impact risk of liquidating positions. Part one calculates the Concentration Threshold or the amount over what can be safely liquidated in 1 day without causing price impact. Any amount over what can be liquidated in one day without price impact is called “Excess Quantity.” Therefore, the concentration charge liquidates a portfolio in a way that aligns with the concentration threshold to ensure no excess quantity exists that causes price impact and toxic debt. # Event Risk The 3AC, FTX, UST, and SVB events show that there are always black-swan-like events that a standard filtered historical simulation VAR risk model doesn’t detect. Concordia’s event risk add-on models worst-case event scenarios for portfolios and applies those returns similar to the price risk model to re-risk the protocol to a black-swan market move. # JSON API Concordia's frontend API for programmatic control Concordia is an API-first platform, that allows for full programmatic & automated control. The documented APIs can be found at the following links: * * * # Margin trading Concordia's collateral protocol presents application developers an easy path to implement margin trading. Using Concordia, one can open trades with borrowed funds, and settle them physically across blockchain networks. Bilbo deposits 100 USDC into his Concordia margin account. He wants to use this deposit to trade on leverage. An on-chain margin-trading program is connected to Concordia, and can perform a **composite operation** which includes a withdrawal, exchange, and deposit all in one transaction. Bilbo executes this program to perform a 5x leveraged swap of USDC for ETH on a connected DEX. * Concordia allows Bilbo to borrow 400 USDC * All 500 USDC are withdrawn from Bilbo’s account * The program performs an exchange for \~$500 worth of ETH * $500 ETH are deposited into Bilbo’s account, and the transaction ends
At the conclusion of this 5x trade, Bilbo is in debt 400 USDC, and owns $500 worth of ETH. His debt is secured by the ETH he bought. Importantly, he performed this 5x trade on an initial deposit of 100 USDC. "Closing" this open position requires paying back the 400 USDC he borrowed. This can be done with a second composite operation, in the reverse: withdraw the ETH, exchange it for USDC, deposit the USDC, and repay the debt.
# Cross-chain money market Concordia can be employed as a universal money market for native digital assets from multiple blockchains. For example, a user may want to borrow APT on Aptos using only ETH on Ethereum as collateral to secure the loan. In this example, they retain their exposure to native ETH (that is, they do not “bridge” it to Aptos). And they gain leveraged exposure to APT on Aptos. They may then spend this APT anyway they please. The cross-chain money market application makes it very easy for users to permissionlessly enter into new blockchain networks, and interact with their applications, without needing to go through a centralized exchange to acquire the native tokens.