<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Amin El-Gazzar on Medium]]></title>
        <description><![CDATA[Stories by Amin El-Gazzar on Medium]]></description>
        <link>https://medium.com/@ensigno?source=rss-1911d31e223d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*u7Vht1GnYtfDGKyjjkSmJg.jpeg</url>
            <title>Stories by Amin El-Gazzar on Medium</title>
            <link>https://medium.com/@ensigno?source=rss-1911d31e223d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 15:39:53 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@ensigno/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[The Probabilistic Paradox: Why LLMs Fail in Deterministic Domains — and How to Fix It]]></title>
            <link>https://medium.com/@ensigno/the-probabilistic-paradox-why-llms-fail-in-deterministic-domains-and-how-to-fix-it-be21b5e20bda?source=rss-1911d31e223d------2</link>
            <guid isPermaLink="false">https://medium.com/p/be21b5e20bda</guid>
            <category><![CDATA[programming]]></category>
            <category><![CDATA[llm]]></category>
            <category><![CDATA[defi]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Amin El-Gazzar]]></dc:creator>
            <pubDate>Thu, 03 Jul 2025 19:01:42 GMT</pubDate>
            <atom:updated>2025-07-03T19:01:42.100Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*EQpRFC5UQtKaBcvPrblH7A.jpeg" /></figure><h3>The Probabilistic Paradox: Why LLMs Fail in Deterministic Domains — and How to Fix It</h3><h3>Introduction: The Probabilistic Paradox</h3><p>Large Language Models (LLMs) have dazzled the world with their capacity to process, generate, and interact with human language at unprecedented scale and fluency. From writing legal memos to summarizing medical records, they seem poised to augment — or even replace — many cognitive tasks. But beneath this promise lies a fundamental tension: <strong>LLMs are probabilistic by nature, while many of the domains they’re being deployed into are deterministic by design</strong>.</p><p>Finance, medicine, law, and engineering are not playgrounds for improvisation. They demand <strong>predictable, auditable, and verifiable decisions</strong> — where the same input must produce the same output, and correctness isn’t just nice to have, it’s mission-critical. And yet, LLMs operate stochastically: even under identical settings, they may produce different answers, vary in formatting, or hallucinate entirely. Recent studies show that state-of-the-art models like GPT-4 exhibit measurable non-determinism across runs, with output variability as high as 70% in certain contexts — even when temperature and prompt are held constant.</p><p>To reconcile this mismatch, developers have tried prompt engineering, chain-of-thought reasoning, and swarms of agents calling external tools. These techniques do improve performance — sometimes dramatically — but <strong>they don’t solve the core problem</strong>: asking a language model, trained on patterns and probabilities, to operate as if it were a rules-based, deterministic system.</p><p>This article explores that contradiction — and proposes a way forward: rather than trying to make the LLM behave deterministically, we should <strong>build deterministic, bounded environments in which the LLM is allowed to act</strong>. Like a chess player constrained by the board and rules, an LLM operating inside a controlled universe can still be flexible, creative, and useful — but without the catastrophic risk of making up moves that aren’t legal.</p><h3>The Incompatibility at the Core</h3><p>At the heart of this issue lies a foundational mismatch between how LLMs work and what deterministic systems demand.</p><h4>LLMs Are Probabilistic Machines</h4><p>LLMs don’t “know” facts or rules — they model language. When asked a question, they don’t consult a database or execute code. They predict, token by token, what is most likely to come next based on patterns observed in massive corpora of human text. This is what gives them their remarkable fluency, but also what makes them unreliable.</p><p>Even under supposedly fixed conditions — same prompt, same model, same temperature — outputs can and do vary. A recent 2024 study quantified this phenomenon across multiple leading LLMs: identical prompts yielded different responses up to 70% of the time, and factual accuracy swung by as much as 15% depending on the sampling. That’s a feature for creativity; it’s a bug for systems that demand reliability.</p><h4>Deterministic Domains Don’t Forgive Variance</h4><p>Deterministic domains, on the other hand, are built on the assumption that <strong>every operation has an expected and verifiable outcome</strong>. In finance, an interest calculation should yield the same result whether run on Tuesday or Sunday. In medicine, a dosage algorithm can’t occasionally “rephrase” its output. In law, precedent isn’t a suggestion.</p><p>The problem isn’t just error — it’s <strong>unpredictable error</strong>. A deterministic system that fails consistently can be debugged. A probabilistic one that occasionally hallucinates a falsehood or forgets a constraint breaks trust completely. That’s why LLMs, left unbounded, are unfit for roles where <strong>repeatability, auditability, and traceability</strong> are not optional — they are mandatory.</p><h4>Simulating Determinism Isn’t the Same as Guaranteeing It</h4><p>To patch this mismatch, practitioners have turned to techniques like prompt engineering (“think step by step”), chain-of-thought reasoning, and agent frameworks that call APIs or tools. These are helpful — they <strong>simulate</strong> more structured, logical thinking — but they do not transform the model’s nature. As studies show, even well-prompted models can skip steps, ignore constraints, or hallucinate intermediate logic. Worse, their reasoning paths — when prompted to explain — often differ from what actually led to the answer.</p><p>In short, you can guide an LLM toward determinism, but you <strong>can’t make it deterministic</strong> by prompting alone. The mismatch is structural, not superficial.</p><h3>Why Band-Aids Aren’t Enough</h3><p>The limitations of LLMs in deterministic domains are well-known, and the community hasn’t been idle. A growing toolkit — prompt engineering, chain-of-thought (CoT) reasoning, agent frameworks, self-consistency, retrieval augmentation — has been developed to push these models toward more structured, reliable outputs. But there’s a problem: <strong>none of these fixes resolve the root issue</strong>. They add scaffolding, not certainty.</p><h4>Prompt Engineering: Art, Not Science</h4><p>Prompt engineering — carefully crafting inputs to guide the model — has become the first line of defense. Want more factual answers? Add “answer concisely and truthfully.” Want reasoning? Add “think step by step.” These tweaks can improve accuracy and format consistency, sometimes dramatically.</p><p>But the gains are <strong>fragile</strong>. A prompt that works on GPT-4 may fail on Claude or LLaMA. Even on the same model, minor wording changes can lead to different answers. Worse, the model may “forget” instructions mid-output — a common failure mode in long or multi-step tasks. Prompting offers influence, not control.</p><p>As one empirical study put it: “Prompt design nudges behavior, but does not guarantee compliance.” There is no hard stop, no formal enforcement. Prompt engineering can help, but it cannot turn a stochastic model into a deterministic one.</p><h4>Chain-of-Thought: Better Reasoning, Still Probabilistic</h4><p>Chain-of-thought prompting pushes LLMs to show their reasoning steps, rather than just the answer. This often improves accuracy in math, logic, and multi-step questions. Adding self-consistency — sampling multiple reasoning paths and aggregating — improves it further.</p><p>But even detailed reasoning can go off the rails. LLMs frequently fabricate or “backfill” explanations that sound logical but don’t actually reflect the internal process. This undermines trust and defeats the purpose of transparency. Moreover, generating multiple reasoning paths is compute-intensive and still doesn’t guarantee correctness — just statistical consensus.</p><p>Chain-of-thought is a <strong>valuable diagnostic tool</strong>, not a deterministic one.</p><h4>Agent Frameworks: Powerful, But Messy</h4><p>Agentic frameworks — where LLMs reason, act, call tools, and iterate — represent the most ambitious effort to bridge the probabilistic-deterministic gap. Projects like ReAct, AutoGPT, and LangChain agents allow models to plan, fetch data, and execute functions with some notion of state and feedback.</p><p>But these systems are <strong>complex and brittle</strong>. Agents often fail in unpredictable ways: using the wrong tool, looping infinitely, or hallucinating the next action. A 2024 study catalogued seven common failure modes in real-world agent deployments — most related to gaps between what the model <em>assumed</em> it had done and what actually happened.</p><p>Without hard constraints and verification, agents don’t solve the determinism problem — they <strong>amplify</strong> it by creating multi-step workflows where each decision can go wrong in a different way.</p><h4>The Common Thread: No Guarantees</h4><p>All of these methods share the same limitation: <strong>they rely on the LLM behaving well</strong>. They steer and scaffold, but they don’t constrain. When they work, it’s because the model “cooperated.” When they fail, there’s no fallback mechanism — and often, no traceable cause.</p><p>In domains that demand predictable outcomes, we can’t afford that kind of ambiguity. The solution is not more clever nudges. It’s a new structure: <strong>a bounded universe in which the LLM must operate</strong>.</p><h3>The Solution: Building Bounded Universes</h3><p>If the problem is that LLMs are fundamentally probabilistic, then the solution is not to make them deterministic — it’s to <strong>design deterministic environments in which they can safely operate</strong>. Rather than trying to force coherence out of chaos, we define the walls of the universe the LLM lives in. Inside that universe, it’s free to reason, interpret, and plan — but the consequences of its actions are bounded, controlled, and verifiable.</p><p>This is the same principle behind secure software sandboxes, strict API contracts, or aircraft flight control systems: <strong>don’t trust the actor — constrain the environment</strong>.</p><h4>What Is a Bounded Universe?</h4><p>A bounded universe is an architectural and operational constraint. It limits:</p><ul><li><strong>What data the LLM can see</strong> (e.g. only retrieved documents, structured inputs),</li><li><strong>What tools or functions it can invoke</strong> (e.g. only approved APIs with strict schemas),</li><li><strong>What outputs it is allowed to generate</strong> (e.g. structured JSON, legal predicates, or code with testable semantics), and</li><li><strong>What happens after it acts</strong> (e.g. human-in-the-loop review, formal verification, or runtime evaluation).</li></ul><p>It’s not about silencing the LLM — it’s about putting it on rails. Within those rails, it can still be creative, adaptive, and language-fluent. But its actions have <strong>bounded variance and deterministic guardrails</strong>.</p><h4>Examples in Practice</h4><p>We’re already seeing early versions of bounded universes emerge in production:</p><ul><li><strong>Finance:</strong> Morgan Stanley’s GPT-4 assistant only generates answers based on retrieved, approved documents. The LLM cannot invent numbers or act beyond what’s in the retrieved context.</li><li><strong>Healthcare:</strong> Experimental tools now combine LLMs with medical knowledge graphs and ontologies, constraining outputs to evidence-based sources, allowing natural language interaction without factual drift.</li><li><strong>Code and Reasoning:</strong> Approaches like PAL (Program-Aided Language models) and LLMLift let LLMs write code or proofs — but require deterministic execution or verification (e.g. running the Python output or checking logic with a theorem prover).</li><li><strong>Security and Planning:</strong> Microsoft’s FIDES system wraps LLM agents with deterministic data flow controls. If the model tries to take an unsafe action (e.g. sending sensitive data), the system blocks it — <strong>not based on trust, but by design</strong>.</li></ul><p>In all these cases, the LLM’s power comes from language and reasoning — but the <strong>certainty comes from the bounded system around it</strong>.</p><h4>From Model to System</h4><p>This shift — from LLM as oracle to LLM as subsystem — is profound. It repositions the model as one component in a broader, <strong>engineered pipeline</strong>. The system’s reliability no longer depends on “did the model behave today?” but on <strong>“did the environment constrain outcomes properly?”</strong></p><p>This is the only realistic way to deploy LLMs into deterministic domains. Not by turning them into something they’re not — but by surrounding them with structure, constraints, and checkpoints that ensure their probabilistic nature doesn’t bleed into mission-critical decisions.</p><h3>Designing for Predictability: A Blueprint for Bounded AI</h3><p>If we accept that probabilistic models alone cannot safely operate in deterministic domains, the question becomes: <em>how do we design systems that harness their strengths without inheriting their weaknesses?</em> The answer lies in treating the LLM not as the system, but as <strong>a subsystem embedded within a deterministic architecture</strong>.</p><p>Here’s a blueprint for doing just that — grounded in real-world use cases and emerging best practices:</p><h4>1. Define the Bounds: What Is the Model Allowed to See and Do?</h4><p>The first step in designing a bounded universe is to clearly define:</p><ul><li><strong>Data Access</strong>: Does the LLM operate purely on a retrieval-augmented corpus? Can it hallucinate facts, or must everything be cited?</li><li><strong>Tool Access</strong>: Can it calculate, look up data, run code? If so, how? What APIs does it use, and who audits them?</li><li><strong>Output Schema</strong>: Does the model return free-form text, or must it conform to a JSON schema, function call, or formal logic representation?</li><li><strong>Postconditions and Enforcement</strong>: Are outputs checked, verified, or executed automatically? Is a human in the loop?</li></ul><p>Think of it as setting the rules of the game: the model can play within them, but cannot redefine them.</p><h4>2. Delegate Determinism to Tools and Verifiers</h4><p>LLMs are excellent at interpretation and synthesis — but terrible at precision. So don’t let them calculate. Don’t let them decide. Let them <strong>delegate</strong> those tasks to deterministic components.</p><p>Examples:</p><ul><li>Financial logic → programmatic rules or spreadsheet-like evaluators.</li><li>Code suggestions → unit-tested before acceptance.</li><li>Legal references → checked against actual clause databases.</li><li>Plans → verified via a formal logic engine or simulation (as in VeriPlan or LLMLift).</li></ul><p>If the LLM proposes something unsafe, invalid, or unverifiable, the system rejects or re-prompts. The burden of correctness shifts to <strong>systems designed for correctness</strong>.</p><h4>3. Use Constrained Interfaces and Structured Decoding</h4><p>Rather than free-text output, define <strong>interfaces</strong>:</p><ul><li>Function calling (e.g. OpenAI tools API)</li><li>Declarative schemas (e.g. JSON with type constraints)</li><li>Controlled natural language (e.g. Prolog rules, SQL queries, or DSLs)</li></ul><p>This makes outputs <em>verifiable</em> and often <em>testable</em>. If the model deviates from the structure, it fails fast — rather than failing silently.</p><h4>4. Treat Verification as a First-Class Citizen</h4><p>Verification is not an afterthought — it’s <strong>a core design layer</strong>. Whether it’s a formal checker, a regression test, a type system, or a human reviewer, every critical system should have <strong>a verification checkpoint before deployment or execution</strong>.</p><p>Think:</p><ul><li>“Does this LLM-generated contract comply with local regulations?”</li><li>“Is this suggested dose consistent with clinical guidelines?”</li><li>“Is this plan executable given our resource constraints?”</li></ul><p>If not — it gets rejected. No exceptions.</p><h4>5. Always Be Able to Say: “We Know What It Did and Why”</h4><p>Transparency matters. In deterministic systems, you can ask: <em>What happened? Why?</em> With LLMs, that’s tricky — but not impossible.</p><p>Design systems to:</p><ul><li>Log every prompt and response.</li><li>Trace decisions to underlying retrieved data or constraints.</li><li>Provide human-readable justifications (chain-of-thought or post-fact explanations).</li><li>Make fallback paths explicit (“We couldn’t validate the model’s plan, so we used the fallback strategy.”)</li></ul><p>Predictability is not just about outputs — it’s about being able to <strong>explain those outputs after the fact</strong>.</p><p>This blueprint is already taking shape in high-stakes deployments. The difference between the systems that succeed in deterministic domains and those that fail is simple: <strong>the successful ones never trust the LLM to act alone</strong>.</p><h3>Case Study: YieldCube — A Bounded Universe for Decentralized Finance</h3><p>To illustrate how a bounded universe can be architected in one of the world’s most demanding deterministic domains, consider the field of Decentralized Finance (DeFi). DeFi operates on public blockchains where transactions are immutable, verifiable, and require absolute precision. There is no room for the kind of probabilistic ambiguity inherent in LLMs. An incorrect action can lead to an irreversible loss of funds.</p><p>This is the exact challenge addressed by the YieldCube architecture, a DeFi operating system that embeds an AI agent within a strictly defined, deterministic environment. It serves as a practical implementation of the blueprint described above.</p><h4>1. Defining the Bounds: The Universal Protocol Gateway</h4><p>The system’s primary interface is the Universal Protocol Gateway (UPG), which sets clear and enforceable limits on the AI agent’s operational environment.</p><ul><li><strong>Bounded Data Access:</strong> The agent is prevented from operating on unverified information or hallucinating facts. Its knowledge is strictly confined to two sources: data read directly from on-chain smart contracts, and information from whitelisted, deterministically provided data streams. This is analogous to real-world, high-stakes deployments where an LLM is only permitted to generate answers based on retrieved, approved documents.</li><li><strong>Constrained Tool Access:</strong> The agent cannot invent actions. Its universe of possible operations is strictly limited to the functions exposed by the UPG, such as deposit, stake, or claimRewards. This ensures every proposed operation is valid within the context of the target protocol.</li><li><strong>Structured Output:</strong> The agent’s intent must be formatted into a structured, machine-readable call (e.g., via MCP or a dedicated API). Free-form text is not permitted for execution, forcing the LLM’s probabilistic reasoning into a deterministic instruction that the system can parse and verify.</li></ul><h4>2. Delegating Determinism: On-Chain Execution</h4><p>YieldCube rigorously follows the principle of delegating tasks to deterministic components. The AI agent does not perform calculations or execute transactions itself. It forms a <em>proposal</em> that is then passed to the blockchain for execution. The Ethereum Virtual Machine (EVM) is a deterministic state machine; it will execute the transaction with formal certainty or fail with a clear, verifiable error. This is analogous to the Program-Aided Language (PAL) model, where the LLM’s output is a piece of code that is then run by a deterministic interpreter to guarantee the result.</p><h4>3. Verification as a First-Class Citizen: The Yobex Guard</h4><p>Before any transaction proposed by the agent is sent to the blockchain, it must pass through a critical verification layer called the Yobex Guard. This component serves as the system’s deterministic guardrail, enforcing rules the LLM cannot be trusted to follow on its own. These guardrails include:</p><ul><li><strong>Role-Based Permissions:</strong> Is the agent authorized to perform this action?</li><li><strong>Policy Enforcement:</strong> Does the proposed transaction violate pre-set rules, such as spending limits or whitelisted contract addresses?</li><li><strong>Timelocks:</strong> Has a mandatory waiting period for a sensitive operation been observed?</li></ul><p>If an agent attempts to generate an unsafe or unauthorized transaction, the Guard deterministically blocks it <em>by design</em>, shifting trust from the probabilistic model to the verifiable environment.</p><h4>4. Auditability and Traceability</h4><p>Finally, every action executed by the system is an on-chain transaction logged on a public ledger. This provides an immutable and fully transparent audit trail, fulfilling the critical requirement that the system can always answer, “We know what it did and why”.</p><p>By embedding its AI agent within this multi-layered bounded universe, the YieldCube architecture successfully harnesses the LLM’s powerful reasoning and planning capabilities for the complex domain of DeFi, without inheriting its dangerous probabilistic weaknesses. It is a clear example of moving from a model we <em>hope</em> will behave to a system we <em>know</em> will.</p><h3>Conclusion: From Language to Logic</h3><p>We’re entering an era where LLMs will increasingly touch high-stakes decisions — not just draft your emails, but analyze medical histories, interpret financial risk, suggest legal strategies, or guide critical infrastructure. These are domains where <em>“close enough” is not enough</em> — where decisions must be traceable, verifiable, and above all, <strong>repeatable</strong>.</p><p>But LLMs aren’t built for that. They’re <strong>statistical instruments</strong>, not deterministic engines. Their power comes from the vastness of their training and the elegance of their fluency — not from formal logic, factual certainty, or adherence to constraint. Trying to make them deterministic through prompting or procedural coaxing misses the point. It’s like trying to turn a jazz musician into a classical pianist by giving them a metronome.</p><p>The solution isn’t to change what LLMs are — it’s to change <strong>how we use them</strong>.</p><p>We don’t ask an intern to launch a rocket. We ask them to take notes, summarize findings, draft proposals — all within the bounds of review and supervision. The same goes for LLMs. Inside a bounded universe — where data is curated, tools are deterministic, outputs are structured, and every action is auditable — LLMs can shine. Outside that boundary, they’re a liability.</p><p>This is the shift that needs to happen:<br> <strong>From LLMs as autonomous decision-makers to LLMs as powerful, bounded collaborators</strong>.<br> From models we <em>hope</em> will behave, to systems we <em>know</em> will.</p><p>Because in a deterministic world, trust isn’t earned by eloquence — it’s earned by <strong>predictability</strong>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=be21b5e20bda" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Eternal Ethereum Bull Strategy]]></title>
            <link>https://medium.com/@ensigno/the-eternal-ethereum-bull-strategy-5dbeb4abf590?source=rss-1911d31e223d------2</link>
            <guid isPermaLink="false">https://medium.com/p/5dbeb4abf590</guid>
            <category><![CDATA[ethereum]]></category>
            <category><![CDATA[bitcoin]]></category>
            <category><![CDATA[defi]]></category>
            <dc:creator><![CDATA[Amin El-Gazzar]]></dc:creator>
            <pubDate>Tue, 25 Jan 2022 15:56:44 GMT</pubDate>
            <atom:updated>2022-01-25T15:56:44.157Z</atom:updated>
            <content:encoded><![CDATA[<p>As many of you know, I have been and still am extremely bullish for Ethereum, even after the brutal slaughter of the last few weeks. Nevertheless, I am convinced that within two years, Ethereum will trade in the low five digits, and within five years, in the high five digits. I don’t have a crystal ball, but I am heavily involved in developing use cases for Ethereum in the DeFi space that will dramatically expand Ethereum usage and price, especially once ETH 2.0 becomes functional.</p><p>But, as many of you also know, I am risk-averse, and volatility makes me nervous. For this reason, most of my crypto investments are in high-yielding stablecoin vaults, returning something between 20–25% annually. While this is a great return, I wanted to increase my exposure to Ethereum while, at the same time, minimizing the potential downside risk.</p><p>And that’s where the Eternal Ethereum Bull Strategy comes in. The goal for the strategy is to allow full or nearly full exposure to Ethereum’s upside price movements while limiting the impact that the inevitable downturns will have on the value of the portfolio. In addition, I want to increase the Ethereum holdings during market pullbacks.</p><p><strong>Sidenote</strong>: <em>I am the CEO &amp; Founder of Yieldster. Yieldster has developed a DeFi automation platform that makes it easy to build automated portfolio management strategies across the entire DeFi space. For the implementation of this strategy, I am using the stablecoin &amp; volatile vaults issued on the Yieldster DeFi Automation Platform, but as the Yieldster Automation Platform has not yet been released to the public, I will suggest alternate stablecoin vaults that produce similar returns but must be maintained manually. This particular strategy is easy to set up and maintain, even if it has to be done manually. Once the Yieldster Platform is opened to the public, we will be releasing this strategy as a vault to the public.</em></p><h3>Strategy Setup</h3><p>The Eternal Ethereum Bull Strategy requires some steps to set up and requires constant maintenance, which means due to the high gas costs on Ethereum these days, this strategy is really only feasible with a portfolio of $100,000 or more. We will start out with $175,000.</p><h3>Step 1:</h3><p>Invest 75% of the portfolio into a high-yielding stablecoin protocol. As mentioned before, I am using a Yieldster stablecoin vault that returns reliably between 20–25% annualized, mostly because it auto-harvests and auto-compounds, which simplifies the daily management tasks. But for the purpose of this article, we can also use the Convex MIM vault, which also returns between 18–25%. However, when using the Convex Vault, the rewards must be manually harvested while also reinvesting the proceeds.</p><h3>Step 2:</h3><p>Use the remaining 25% of the portfolio to buy Ethereum. At the time I was setting up the strategy, December 9, 2021, I bought 12 ETH for $4,319.20, or roughly $50,000. As you can probably see, my market timing isn’t too great, as immediately after I bought the Ethereum, the price collapsed by almost 25% over the next three weeks. But, that’s ok. The strategy is designed to handle downturns better than just holding ETH while using retracting ETH prices to increase ETH holdings.</p><h3>Step 3:</h3><p>Deposit the 12 ETH to Liquity and borrow LUSD up to a Debt to Asset ratio of 150% (or ~67% of the asset value. At the time I set up this strategy, I was able to borrow 33,400 LUSD against the 12 ETH.</p><h3>Step 4:</h3><p>Deposit the 33,400 LUSD into the Convex Tricrypto2 Vault on Abracadabra and borrow MIM against it, again up to a Debt to Asset ratio of 150%. In this case, I was able to borrow 23,000 MIM against the collateral. We are using the Curve Tricrypto2 basket for this step as it acts as a dampener to the sometimes extreme volatility of Ethereum while at the same time still offering great exposure to the potential upside. I, again, used a Yieldster Volatile Vault based on Convex Trycrypto2 as it automates harvesting and compounding, but the returns are very similar to investing directly into the Convex Vault. We expect to earn about 20–25% on our Convex Trycrypto holdings.</p><p>The MIM loan will cost an upfront fee (currently 0.5%) plus ongoing interest of currently 3.5%.</p><h3>Step 5:</h3><p>Deposit the 23,000 MIM into the same stable coin vault you invested into in step 1.</p><h3>Portfolio After Setup:</h3><ul><li>Stablecoin Vault total assets: $147,800</li><li>Earning 20–25% annually or $30–37k annually</li><li>TryCrypto2 total assets: 18.74TryCrypto2 worth ~$33,400</li><li>Earning 20–25% annually or $6.7–8.5k annually</li><li>ETH total assets: 12 ETH worth ~$50,000</li><li>Liquity LUSD Debt: $33,400</li><li>No interest loan</li><li>Abracadabra MIM debt: $23,000</li><li>3.5% annual interest</li></ul><p>In summary, we have a portfolio with $231,200 in assets and $56,400 in debt for a net portfolio value of $174,800.</p><ul><li>$50,000, or 25% have full exposure to ETH price movements</li><li>$33,400 or 19% have partial exposure to ETH and BTC price movements</li><li>In total roughly 44% have exposure to ETH ( and BTC) volatility</li></ul><h3>Strategy Maintenance</h3><p>This strategy requires some maintenance after the setup:</p><h3>Harvesting and Compounding</h3><p>If you are not using Yieldster Stablecoin and Volatile Vaults to set up the strategy, you will have to harvest the rewards earned yourself and reinvest in the underlying vaults. The Yieldster vaults automatically optimize harvesting rewards against gas costs so you don’t have to worry about it. But, if you are running the strategy manually, we suggest harvesting once the accumulated rewards are $5,000 or more and wait for low gas costs. Harvesting frequency and timing are not critical to this strategy, so minimizing gas costs should be the main focus here.</p><p>All harvested rewards are sold and invested back into the Yieldster Stablecoin or Convex MIM vault.</p><h3>Maintaining Debt-to-Asset Ratios</h3><p>As we are using leverage in this strategy. we need to closely monitor the debt-to-asset ratios for our Liquity loan and the Abracadabra loan. When we set up the strategy, we borrowed up to 150% debt-to-asset ratio for both loans. As a precaution, we will be adding collateral every time the debt-to-asset ratio drops below 140%. Ultimately, our goal is to make sure that our debt-to-asset ratio always ranges between 140–150%.</p><p>When the debt-to-asset ratio drops below 140%, we will be using the stablecoin holdings to buy more collateral until the debt-to-asset ratio is again over 150%. If the debt-to-asset ratio is more than 150%, we will be borrowing more LUSD or MIM and depositing into the stablecoin vault or volatile vault, respectively.</p><p>Again, the Yieldster Automation Platform automatically optimizes for gas costs so you don’t have to worry about it. If you are running the strategy manually, we suggest setting a minimum transaction amount to prevent excessive and expensive transactions. For the portfolio size outlined here, we suggest a minimum of $5,000.</p><h3>Maintaining Liquity LUSD debt-to-asset Ratio</h3><h4>Debt-to-Asset ratio drops below 140%</h4><p>If the debt-to-asset ratio drops below 140%, we will buy ETH and deposit it as additional collateral into the Liquity Trove. The amount of ETH we buy will be the larger of $5,000 or the amount needed to change the debt-to-asset ratio to over 150%.</p><h4>Debt-to-asset ratio grows beyond 150%</h4><p>If the debt-to-asset ratio grows beyond 150% we will borrow LUSD against the collateral until the debt-to-asset ratio reaches again 150%. We will only borrow additional LUSD if we can borrow a minimum of at least $5,000 before the debt-to-asset ratio drops to 150% after the additional loan</p><p>In effect, we will be increasing our ETH holdings when ETH prices are going down and we will increase our stablecoin holdings when ETH prices are going up.</p><h3>Maintaining Abracadabra MIM debt-to-asset Ratio</h3><h4>Debt-to-Asset ratio drops below 140%</h4><p>If the debt-to-asset ratio drops below 140%, we will add collateral to the Abracadabra Convex Trycrpto2 vault using our stablecoin holdings. The amount of TryCrypto2 we buy will be the larger of $5,000 or the amount needed to change the debt-to-asset ratio to over 150%.</p><h4>Debt-to-asset ratio grows beyond 150%</h4><p>If the debt-to-asset ratio grows beyond 150%, we will borrow MIM against the collateral until the debt-to-asset ratio reaches 150% again. We will only borrow additional MIM if we can borrow a minimum of at least $5,000 before the debt-to-asset ratio drops to 150% after the additional loan.</p><p>Again, in effect, we will be increasing our TryCrypto2 holdings when ETH &amp; BTC prices are going down and we will increase our stablecoin holdings when ETH &amp; BTC prices are going up.</p><h3>Strategy Performance</h3><p>We opened this strategy on December 9, 2021, at an Ethereum price of $4,319.20, and at the time of writing, January 24, 2022, the price of Ethereum has crashed to $2443.89, for a total loss of more than 43%. If we would have invested $175,000 in ETH, our portfolio would be now worth a little under $101,700. The Eternal Ethereum Bull strategy, on the other hand, “only” dropped by 22.4% during the same time for a net value of just under $137,000. At the same time, we increased our holdings of ETH from 12 ETH to 27.7 ETH (+131%) and TryCrypto2 from 18.74 to 27.46 (+47%)</p><p>The below chart effectively shows how this strategy protects against the downside while at the same time participating almost fully on the upside moves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/977/0*PKc0gSzHUGSns5Fv.jpg" /></figure><h3>Strategy Portfolio On January 24th, 2022:</h3><ul><li>Stablecoin Vault total assets: $98,200</li><li>Earning 15–20% annually or $15–20k annually</li><li>TryCrypto2 total assets: 27.46 triCrypto2 worth $35,000</li><li>Earning 10–15% annually or $3–5k annually</li><li>ETH total assets: 27.7 ETH worth $69,000</li><li>Liquity LUSD Debt: $43,400</li><li>No interest loan</li><li>Abracadabra MIM debt: $23,065</li><li>3.5% annual interest</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/977/0*QrJl7h4YHMOM1LyO.jpg" /></figure><p>In summary, the portfolio holds assets worth $203,000 with $66,500 in debt for a net portfolio value of $136,500. However, this is still substantially better than investing the full $175,000 in ETH on December 9th, the portfolio would now (January 24th, 2022) only be worth $102,000.</p><h3>Outlook</h3><p>I was hoping to write this article while Ethereum continues to grind its way higher, but that is taking too long. I continue to be bullish on Ethereum, and I will continue with the strategy.</p><p>The strategy was designed to protect against some of the downside while creating exposure to the upside. And that’s been working like a charm, at least on the way down…</p><p>Now the question is, will it work on the way up as well? Will there ever be a way up? Is the sky falling?</p><p>I’ll be posting regular updates…</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5dbeb4abf590" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Push to Build a DeFi Safe Space for Institutions and Regulated Investors]]></title>
            <link>https://medium.com/yieldster-dao/the-push-to-build-a-defi-safe-space-for-institutions-and-regulated-investors-dd7b239b65e0?source=rss-1911d31e223d------2</link>
            <guid isPermaLink="false">https://medium.com/p/dd7b239b65e0</guid>
            <category><![CDATA[ethereum]]></category>
            <category><![CDATA[defi]]></category>
            <category><![CDATA[investing]]></category>
            <category><![CDATA[cryptocurrency]]></category>
            <dc:creator><![CDATA[Amin El-Gazzar]]></dc:creator>
            <pubDate>Wed, 21 Jul 2021 00:00:09 GMT</pubDate>
            <atom:updated>2021-07-21T00:00:09.304Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*xv0jx3AEenjoaNKC-TZdwA.jpeg" /><figcaption>Creating a DeFi Safe Space for Institutional Investors</figcaption></figure><p>Decentralized finance (DeFi) is growing. People are tired of the regulations tied up with traditional banks. They’re tired of watching their money be lost to fees and expensive overhead costs.</p><p>Regulated investors want a better space. Institutions, too, want a better space.</p><p>DeFi offers the solutions to some of the most modern problems that pertain to banking and investing in traditional manners.</p><p>Creating a DeFi space makes it possible to tap into the explosive growth within the industry. The growing interest in digital assets is proof that institutional investors are ready to take the plunge.</p><p>The DeFi space has already seen tremendous growth — from $700 million to nearly $80 billion in a single year. And within the next few years, it could be substantially more as people move trillions from traditional and centralized institutions to DeFi.</p><p><strong>Why is the Interest Growing?</strong></p><p>Many people are interested in digital assets, such as cryptocurrencies, blockchain tokens, and more. So much of our world has already gone digital. It’s not a passing fad. Now that this has become universally accepted, digital assets are becoming more accepted — and more lucrative.</p><p>A report published by Fidelity Asset Management in early 2021 identified that 80% of institutions surveyed have expressed an interest in investing in digital assets. 36% have already made the plunge in one way or another, with Bitcoin being the asset of choice.</p><p>Evertas, a cryptocurrency insurance firm, has already seen that 90% of those surveyed in the US and UK already plan on increasing their cryptocurrency investments.</p><p>Why are institutional investors looking to move to digital assets? The benefits are simple:</p><ul><li>Lower transportation costs</li><li>Lower transaction costs</li><li>Lower storage costs</li></ul><p>When the costs of acquiring and maintaining digital assets are significantly lower than real estate or hedge funds, it makes sense.</p><p>In the past, digital assets have been out of reach for many because of not having sufficient regulation, untrusted custodians, or a lack of insurance.</p><p>Hastily developed and released protocols in the DeFi space led to a few hundred million being lost to hackers and economic exploits, but even with these staggering numbers the thefts still amount to only a fraction of a percent of the money under management in DeFi. This only highlights the need for regulation and trusted custodians. Otherwise, digital assets are stored based on hopes and dreams.</p><p>Institutional investors want to see trusted companies stepping up to offer custody. They want to have confidence, peace of mind, and a familiar operating model. The New York Department of Financial Services for example stepped in to charter a trust company to secure their digital assets. This solution provides the needed “safe space” so that pension funds, foundations, endowments, and other institutional players have the peace of mind they so desperately need.</p><p>Bitcoin and other digital assets are gaining popularity, but they haven’t gone mainstream among investors yet. Why? Predominantly due to a lack of insurance, regulation, and security.</p><p>By creating a DeFi safe space that offers insurance, regulation, and security, it can ensure that institutional investors can deploy real money. It allows digital assets to gain more momentum.</p><p>More platforms are popping up to help institutional investors gain confidence in the digital asset market. When the same benefits are offered in the traditional world of securities, it provides greater peace of mind and ensures that institutional investors feel comfortable moving into digital assets with the same amount of money that they would dedicate to hedge funds and other traditional alternative asset classes.</p><p>The Yieldster Project is unique because it provides institutional investors with protocols to participate in DeFi with greater confidence. They can stay within their regulatory frameworks while taking advantage of the tremendous and stable returns through automation and streamlining. Additionally, the Yieldster platform offers access to tokenized real-world income streams eliminating the reliance on the often volatile crypto space.</p><p><strong>Legislation and Regulation is Changing and Adapting to Digital Assets</strong></p><p>For years, the SEC was content to ignore what was going on with digital assets. They assumed that it was a temporary disruption and that it would pass. However, it’s clear that such things as cryptocurrencies and tokens are here to stay. And with that realization regulators will have no choice but to step in and create ever more stringent regulations to minimize the risks for investors</p><p>With the shift in legislation comes new regulations as well as new opportunities for insurance. And some innovative products are already on the horizon.</p><p>The seriesOne Liva Fund, for example, which is built on top of the Yieldster platform is a fully licensed and regulated DeFi investment fund that can take advantage of the many opportunities the DeFi space offers for institutional investors and wealth managers. It’s the first time that a regulated fund that uses automated DeFi investment strategies was allowed to enter the regulated investment markets.</p><p>Even the SEC is becoming more tolerant of public blockchain investment vehicles. And with the SEC on board, it offers more confidence to institutional investors who have hung back from taking the plunge into digital assets.</p><p>As the interest grows, it’s not just about the digital assets themselves. It’s also about DeFi protocols and new assets and asset classes moving into the DeFi space. Unlike investing in volatile crypto assets like Bitcoin or Ethereum, the DeFi space enables investors and money managers to design products that for all practical purposes eliminate most of the volatility inherent in the crypto space.</p><p>Is there more risk? Not necessarily. While there is risk in investing in DeFi these risks are on a different level compared to traditional alternative asset classes and many of these risks can be mitigated away. And that’s why the Yieldster Project has been developed.</p><p><strong>The Future of DeFi and Traditional Investments</strong></p><p>More companies are seeing the need to embrace DeFi just as much as they embrace traditional investments. It’s leading to the desire to make the DeFi space safer through protocols, insurance, and smart contract safeguards.</p><p>When risks can be mitigated, it can demand more interest from traditional investors.</p><p>Traditional investments are moving into DeFi. It adds diversity to the ecosystem and allows people to gain success from the cryptocurrency market. As more people adopt crypto, it leads to higher liquidity and reduced volatility.</p><p>There are challenges, yet all things require time. With greater interest comes the demand for a DeFi safe space that can provide the solutions.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=dd7b239b65e0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/yieldster-dao/the-push-to-build-a-defi-safe-space-for-institutions-and-regulated-investors-dd7b239b65e0">The Push to Build a DeFi Safe Space for Institutions and Regulated Investors</a> was originally published in <a href="https://medium.com/yieldster-dao">Yieldster DAO</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Four out of Five Institutional Investors Expect to Increase Their Crypto Exposure Over the Next Two…]]></title>
            <link>https://medium.com/yieldster-dao/four-out-of-five-institutional-investors-expect-to-increase-their-crypto-exposure-over-the-next-two-887ccc105d6a?source=rss-1911d31e223d------2</link>
            <guid isPermaLink="false">https://medium.com/p/887ccc105d6a</guid>
            <category><![CDATA[crypto]]></category>
            <category><![CDATA[defi]]></category>
            <category><![CDATA[cryptocurrency-investment]]></category>
            <dc:creator><![CDATA[Amin El-Gazzar]]></dc:creator>
            <pubDate>Sat, 17 Jul 2021 19:14:49 GMT</pubDate>
            <atom:updated>2021-07-17T19:14:49.549Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong>Four out of Five Institutional Investors Expect to Increase Their Crypto Exposure Over the Next Two Years</strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*BDVBKFLLQk102vqQRqRiBg.jpeg" /><figcaption>Institutional Investors Planning to Increase Crypto Holdings</figcaption></figure><p><strong>Two out of five are increasing their holdings dramatically</strong></p><p>Crypto exposure is about to expand — or at least that’s what <a href="https://www.institutionalassetmanager.co.uk/2021/07/06/302995/study-shows-institutional-investors-are-increasingly-focusing-cryptocurrencies">82 percent of institutional investors and wealth managers</a> believe. More and more people are eager to expose their investments to the crypto market because of the technology and the new opportunities.</p><p>Between now and 2023, investors from the US, France, Germany, the UK, and the UAE are planning to increase their holdings — with 40% saying that they will do so “dramatically.”</p><p><strong>Hopes &amp; Dreams in the Crypto Market</strong></p><p>Cryptocurrencies are new and exciting. With various virtual currencies hitting the news, such as Bitcoin and Dogecoin, investors want to explore the possibilities. Nickel Digit Asset Management conducted a survey among investors to find out their thoughts on virtual asset exposure. Over the next two years, 80% plan on increasing their investments in the crypto markets.</p><p>One of the top reasons for investing in such markets is the long-term growth that is promised. Over 50% of investors are excited about the prospects. Further, people have claimed that they are gaining confidence in this asset class — and that many will allocate funds into the crypto market if they see a giant corporation doing so.</p><p>Cryptocurrencies are becoming more mainstream. People are embracing it as a form of currency — and new coins are coming online all the time. As more corporations endorse the market as an acceptable investment stream, it will lead to more confidence and more exposure.</p><p><strong>A New Generation of Investors</strong></p><p>Digital assets are unique. As such, it’s creating interest among a new generation of investors — including millennial millionaires. A survey conducted by the Spectrum Group identified that some of the biggest investors in the crypto market are the wealthy millennials with portfolios of $1 million +.</p><p>Approximately 50% of this demographic had crypto encompassing at least 25% of their portfolio. Meanwhile, at least 30% had dedicated at least half of their portfolio to it.</p><p>There’s a good reason why the younger generations find the market so alluring, according to George Walper, the Spectrum Group’s president. They were “intellectually engaged” with the idea, even before cryptocurrencies became mainstream.</p><p>With such exciting opportunities and with more people learning about crypto markets every day, it’s only natural to see more market exposure in the next several years.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=887ccc105d6a" width="1" height="1" alt=""><hr><p><a href="https://medium.com/yieldster-dao/four-out-of-five-institutional-investors-expect-to-increase-their-crypto-exposure-over-the-next-two-887ccc105d6a">Four out of Five Institutional Investors Expect to Increase Their Crypto Exposure Over the Next Two…</a> was originally published in <a href="https://medium.com/yieldster-dao">Yieldster DAO</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>