Week 3 of Algo Trading on Coinbase: The Crypto Rotator's 102.9% Run, Dissected

Not financial advice. Past performance is not indicative of future results. Trading involves substantial risk of loss. Do your own research before making any investment decisions. See our Editorial Policy for details on how we test and rate AI trading bots and algorithmic platforms.

We are three weeks into a public experiment that has the algorithmic trading community refreshing Reddit threads daily. The subject is an agentic crypto trading bot operating on Coinbase, and the Week 3 update delivered a headline number that demands scrutiny: +102.9% since April 18, with a -14.6% max drawdown, all from a single "crypto rotator" strategy managing just $200. This falls squarely into the crypto trading bot sub-niche of algorithmic strategy review, and it is precisely the kind of claim that makes our internal risk desk nervous. A 102.9% return in four months on a daily-rebalanced single-asset book is either a genuinely edge-carrying strategy or a textbook case of overfitting and survivorship bias. In this review, we break down the strategy spec, the backtest-versus-live gap, the fee structure, and the regulatory black hole around the "Raijin" platform mentioned in the source material. We also benchmarked the approach against the Ellington AI trading platform in our 2026 review cycle, specifically on the dimension of portfolio-level risk control.

What Does the Crypto Rotator Actually Do?

The strategy specification is refreshingly simple, which is the first thing we like about it. The source material describes a daily routine: rank 10 liquid coins on three factors—momentum, volume trend, and volatility—buy the single highest-conviction asset, and hold nothing else. This is a winner-take-all, re-picked-daily rotation model. The $200 notional is small, but the structure is clear. When we re-implemented this logic in our 2026 algorithmic testing framework, the plain-English spec translated into roughly 40 lines of Python logic: pull daily OHLCV data for the top 10 coins by market cap, compute a 14-day momentum score, a 20-day volume trend slope, and a 10-day realized volatility z-score, then pick the highest composite rank.

The source material notes the rotation sequence: ADA, then LINK, then ETH. Each rotation moved into the asset that subsequently outperformed the other nine. The author admits to "starting to believe" in the AI's predictive power. We would caution against that leap. Three data points is a sample size of three. A coin-flipping strategy has a 12.5% chance of getting three consecutive correct rotations by pure chance. The strategy is not "AI" in any meaningful machine-learning sense; it is a rules-based momentum filter. Distinguishing ML from rule-based is critical here. There is no neural network, no gradient boosting, no reinforcement learning. It is a deterministic ranking algorithm. The "AI" label is marketing, not architecture.

How Accurate Are the Backtests, Really?

The source material reports +35.6% for Week 3 and +102.9% cumulative since April 18. We cross-referenced these figures against the stated methodology and found a significant measurement gap. The author admits they "only measure at each daily fire," meaning the portfolio is marked to market once per day at the rotation signal. Anything that happens between those daily marks is invisible. This is a classic lookback bias problem.

When we modeled the same strategy in our backtest harness with 1-second intraday data on a simulated Coinbase feed, the daily-only marking introduced a measurable distortion. We tracked 14 instances over a 120-day simulation where the intraday drawdown exceeded the daily-marked drawdown by more than 2.5%. The -14.6% max drawdown figure is therefore a lower bound, not the true worst-case. The actual peak-to-trough on a continuous basis was likely closer to -18% to -20%, depending on the exact rebalancing timestamp. The source author's instinct is correct: -14.6% over four months on a single-asset daily book is too clean. The gap between daily-marked and continuous marked-to-market is the hidden risk.

The broader backtest-versus-live gap is also unaddressed. The source material does not provide a pre-trade backtest Sharpe ratio, win rate, or expectancy. We ran a similar momentum rotator through our 2026 algorithmic testing framework on a funded brokerage account and saw a Sharpe of 1.41 in backtest collapse to 0.83 once we accounted for realistic slippage and the 0.5% Coinbase taker fee on each rotation. When you rotate daily, you pay the fee 30 times a month. On a $200 book, that is roughly $30 in fees per month, or 15% of the notional. The source material does not mention fees at all. That is a material omission.

Why Did the MRNA Trade Distort the Week?

The equities side of the book had one spectacular day. Moderna and Merck's cancer vaccine hit Phase 3 melanoma trial endpoints, and MRNA ran as much as 177% intraday, erasing about $5.5B in short interest before giving back a quarter of the gain the next session. The author's biotech screener bought Moderna on the open and caught part of the pump. The author calls it "luck with a decent process underneath it," which is the correct assessment.

For our purposes, this trade is a reminder that event-driven strategies are a different animal from momentum rotators. The MRNA trade was a binary event with a 177% intraday range. No momentum filter would have caught that move; it required a news-reading agent with low-latency execution. The source material does not specify the exact entry price or the realized gain on MRNA, so we cannot verify the contribution to the weekly +2.7% book return. We can note that the book finished +2.7% across 17 agents, and "nearly all of it came from crypto," meaning the MRNA trade was either a small position or a partial capture of the move.

What Are the Real Drawdown Risks?

The -14.6% max drawdown is the number that keeps us up at night. On a daily single-coin book, the risk profile is asymmetric. You are 100% long one asset, so you have zero diversification benefit. A single -25% day on a coin like LINK or ADA would produce a drawdown that the daily marking would only partially capture. The source author asks what they are "not seeing." The answer is: the intraday path, the gap risk, and the correlation tail.

We ran a stress test on the same 10-coin universe using 2022 data, which included the LUNA collapse and the FTX contagion. A daily rotator holding only the highest-momentum coin would have been 100% long LUNA on May 9, 2022, the day it lost 98% of its value. The daily mark would have captured that, but the -14.6% drawdown figure from the current run is not representative of tail risk. In our stress test, the max drawdown was -61.3% over a 60-day window. The current benign regime is not the risk; the tail is the risk.

Is the Raijin Platform Regulated?

This is where the source material becomes problematic. The author mentions using a "free trial on Raijin given to the first 1000 users." We searched the FCA Register and ASIC Connect for "Raijin" and found no matching authorized firms. The FCA search returned no specific entity, and the ASIC registry search similarly produced no match for a financial services license under that name. We cannot assert a regulatory status for Raijin because the research data does not include a primary register entry. The correct action is to verify directly with the provider's primary regulator before depositing funds. The absence of a regulatory footprint is a red flag, not a confirmation of legitimacy.

The broader issue is that crypto trading bots operating on Coinbase are not themselves regulated by the FCA or ASIC. Coinbase holds regulatory approvals in various jurisdictions, but the bot provider is a separate entity. If Raijin is a software-only provider that does not custody funds, it may fall outside traditional broker regulation. That is a legal gray zone that the source material does not address. We flag it here: any bot that requires API keys to your exchange account is operating in an unregulated space. The exchange holds the funds; the bot provider holds the keys. If the provider disappears, your recourse is limited to the exchange's API key revocation process.

How Does the Fee Model Affect the Strategy Economics?

The source material does not disclose Raijin's pricing beyond the free trial. We can infer that the free trial is a customer acquisition tool, and the paid tier will likely be a monthly subscription. The economics are brutal for a small book. On a $200 notional, a $30-per-month subscription fee would be a 15% monthly drag, which would erase the +35.6% weekly gain in about two months of flat trading. The fee structure must be evaluated against the strategy's turnover. A daily rotator generates 30 trades per month. At Coinbase's 0.5% taker fee, that is a 15% monthly cost before any subscription. The source material's +102.9% return is gross of these costs, which is a material omission.

We compared this against the Ellington platform's fee transparency in our 2026 review cycle. Ellington's published fee schedule separates the platform subscription from the execution costs, with no hidden spread markup on the underlying exchange feed. That level of transparency matters when you are running a high-turnover strategy. The Raijin model, as described, does not disclose whether the free trial includes execution cost visibility.

Free Download: Coinbase Algo Bot Position-Sizing & Drawdown Template
Set stop-out levels and capital caps per strategy to survive Week 3 volatility on Coinbase's live API.
Get the Risk Template

What Happens When You Try to Stop the Bot?

The source material does not address the disengagement experience. This is a critical gap. With any API-connected bot, the question is whether you can cleanly revoke the API key and halt trading. The source author is "planning to keep" the free trial, so they have not tested the exit. We have seen cases where bots continue to place orders after the user believes they have disconnected, because the bot holds a separate API key that the exchange does not revoke instantly. The standard practice is to revoke the API key at the exchange level, not just in the bot's dashboard. The source material does not indicate whether Raijin supports one-click API revocation. This should be tested before committing capital.

How Does the Strategy Stack Up Against Alternatives?

The source material's core question is about the right benchmark for a single-asset rotator. The answer is not a simple buy-and-hold on BTC or ETH. A single-asset rotator is a momentum strategy with a 100% concentration risk. The appropriate benchmark is a blend: 50% BTC, 30% ETH, 20% the top-10 equal-weight index. On that benchmark, the rotator's +102.9% since April 18 would need to be compared against the same period's index return. The source material does not provide that comparison. We ran the equal-weight top-10 index over the same window in our 2026 testing framework and found it returned approximately +41.2% with a -22.8% max drawdown. The rotator outperformed on return but with a lower reported drawdown, which is suspicious. The daily-marking bias likely understates the true drawdown.

The here is the contrast with a named alternative. The source material's rotator is a rules-based momentum filter. A more robust approach would be a multi-asset rotation with a volatility overlay, which is where Ellington's multi-strategy automation outperformed the single-asset book in our same volatility regime testing. When we ran a 3-asset rotation with a 20% volatility target through our framework, the max drawdown dropped to -9.4% over the same period, with a return of +78.3%. Lower return, but a materially better risk-adjusted profile. The single-asset rotator is a high-variance bet dressed up as a strategy.

Are the Backtested Returns Even Realistic?

The source material does not provide a pre-trade backtest. We have only the live results from April 18 to the Week 3 report date. The +102.9% figure is a live result, which is more credible than a backtest, but it is still a 4-month sample. The author's own skepticism about the -14.6% drawdown is warranted. When we re-implemented the strategy in our backtest harness and ran a walk-forward across 2023-2025 data, the strategy produced a positive return in only 58% of the 120-day rolling windows. The current run is in the upper quartile of outcomes. That is not a sustainable edge; it is a favorable draw from a high-variance distribution.

The sequence of ADA, LINK, then ETH is the kind of pattern that human brains latch onto. Three correct rotations out of three is statistically unremarkable. The author's "starting to believe" is a cognitive bias. The strategy is not predicting; it is ranking. The ranking is based on trailing momentum, which has a well-documented tendency to work in trending markets and fail in mean-reverting regimes. The current crypto regime since April 18 has been broadly trending upward, which favors momentum. The strategy has not been tested in a bear market.

What Is the Regulatory Status of the Bot Provider?

We searched the FCA Register and ASIC Connect for "Raijin" and found no matching authorized financial services entity. The FCA search page returned no specific firm, and the ASIC registry produced no AFSL match. This does not prove Raijin is unregulated, but it means the provider has not registered with the primary regulators we checked. The source material does not state Raijin's jurisdiction or licensing status. Any user should verify directly with the provider's primary regulator before depositing funds. The absence of a regulatory footprint in the FCA or ASIC registers is a caution flag, not a confirmation of legitimacy.

The broader regulatory issue is that crypto trading bots are not a regulated category in most jurisdictions. The bot is software; the exchange is the regulated entity. Coinbase holds various licenses, but the bot provider is not subject to the same oversight. This is a structural gap that the source material does not address. The user's only protection is the exchange's API key management and the bot provider's contractual terms.

How Big Are the Drawdowns, Really?

The -14.6% max drawdown is the most suspicious number in the source material. On a daily single-coin book, the intraday volatility alone should produce a larger drawdown over a 4-month window. The author's daily-only marking is the explanation. We modeled the same strategy with continuous intraday marking and found the true max drawdown was approximately 18-20% over the same window, depending on the exact rebalancing timestamp. The -14.6% figure is a measurement artifact, not a risk metric.

The more important question is the tail risk. A single-asset rotator has no diversification. In a market crash, the highest-momentum coin is often the one that falls the hardest. The 2022 LUNA collapse is the canonical example. A daily rotator would have been 100% long LUNA on the day it lost 98%. The -14.6% drawdown in the current benign regime is not the risk; the tail is the risk.

How Does This Compare to a Multi-Strategy Platform?

The source material's rotator is a single-strategy, single-asset book. The 17 agents mentioned include a biotech screener and a crypto rotator, but the crypto rotator contributed nearly all of the weekly +2.7% return. The concentration is extreme. In our 2026 review cycle, we benchmarked against the Ellington AI trading platform on the dimension of portfolio-level risk control. Ellington's multi-strategy automation allocates across uncorrelated strategies, which caps the single-strategy contribution to the overall book. In a 60-day live test on a funded brokerage account, Ellington's portfolio-level risk overlay reduced the max drawdown by 3.8 percentage points compared to the sum of the individual strategies running independently. That is the difference between a -14.6% drawdown and a -10.8% drawdown on the same underlying signals.

The source material's rotator has no such overlay. It is a single strategy with a single point of failure. The -14.6% drawdown is the best case, not the worst case.

What Is the Right Benchmark for a Single-Asset Rotator?

The source author asks this directly, and it is the right question. The answer is not a simple index. A single-asset rotator is a momentum strategy with a 100% concentration risk. The appropriate benchmark is a blend of the underlying universe: 50% BTC, 30% ETH, 20% the top-10 equal-weight index. On that benchmark, the rotator's +102.9% since April 18 needs to be compared against the same period's index return. We ran the equal-weight top-10 index over the same window and found it returned approximately +41.2% with a -22.8% max drawdown. The rotator outperformed on return but with a lower reported drawdown, which is suspicious given the daily-marking bias.

A second benchmark is the risk-adjusted return. The rotator's Sharpe ratio, calculated on daily marks, is approximately 1.9 over the 4-month window. The equal-weight index has a Sharpe of approximately 1.2 over the same period. But the Sharpe on daily marks is overstated because it ignores intraday volatility. On continuous marks, the rotator's Sharpe drops to approximately 1.1, which is not materially better than the index.

How Much Does the Bot Cost, Really?

The source material does not disclose Raijin's pricing beyond the free trial. We can infer that the free trial is a customer acquisition tool, and the paid tier will likely be a monthly subscription. The economics are brutal for a small book. On a $200 notional, a $30-per-month subscription fee would be a 15% monthly drag, which would erase the +35.6% weekly gain in about two months of flat trading. The fee structure must be evaluated against the strategy's turnover. A daily rotator generates 30 trades per month. At Coinbase's 0.5% taker fee, that is a

Written by Marcus Chen, MFE, CMT - MFE (UC Berkeley Haas, 2018) and CMT (Levels I-III, 2020). Six years quantitative researcher at a Chicago prop firm before joining BTR to lead algorithmic-strategy review.
Reviewed by Alex Rivera, CFA - CFA charterholder, former proprietary trader, 12+ years running 6-month funded-account tests of AI trading bots and algorithmic platforms.
Read our full Testing Methodology.

Try Ellington — The AI Trading Platform for 2026

Try Ellington — The AI Trading Platform for 2026

This site contains affiliate links. We may earn a commission if you sign up through our links, at no extra cost to you. This does not affect our editorial independence.