Cryptocurrency#
Cryptocurrency markets exhibit distinctive properties that set them apart from traditional financial markets: extreme volatility, 24/7 trading without circuit breakers, rapid market microstructure evolution, fragmentation across exchanges, and a participant base that mixes retail, institutional, and algorithmic traders. Understanding these properties is essential for building strategies that are grounded in the actual mechanics of crypto markets rather than naive applications of equity or futures frameworks.
The research here covers foundational crypto asset pricing — how crypto risk factors are structured, what drives cross-sectional return differences, and how systemic risk propagates across the ecosystem. Topics include the empirical distribution of crypto returns, low-volatility anomalies, factor models tailored to crypto, cryptocurrency contagion dynamics, and the practical implications for institutional investors entering the space.
Related topics include DeFi & AMMs for on-chain market mechanics, Carry Trade for funding rate strategies, Momentum for crypto momentum strategies, and Order Flow for microstructure-driven approaches.
For complete worked examples of trading strategies and individual factors, see Strategy Examples and Factor Examples.
An Investor’s Guide to Crypto#
We provide practical insights for investors seeking exposure to the growing cryptocurrency space. Today, crypto is much more than just bitcoin, which historically dominated the space but accounted for just a 21% share of total crypto trading volume in 2021. We discuss a wide variety of tokens, highlighting both their functionality and their investment properties. We critically compare popular valuation methods. We contrast buy-and-hold investing with more active styles. We only deem return data from 2017 representative, but the use of intraday data boosts statistical power. Underlying crypto performance has been notoriously volatile, but volatility-targeting methods are effective at controlling risk, and trend-following strategies have performed well. Crypto assets display a low correlation with traditional risky assets in normal times, but the correlation also rises in the left tail of these risky assets. Finally, we detail important custody and regulatory considerations for institutional investors.
Low-volatility strategies for highly liquid cryptocurrencies#
Managing extreme price fluctuations in cryptocurrency markets are of central importance for investors in this market segment. Using a sample of highly liquid cryptocurrencies from January 2017 to June 2021, this paper proposes a dynamic investment strategy that selects cryptocurrencies based on their historical volatility and is complemented by a simple stop-loss rule. Our results reveal that investing in highly concentrated low volatility cryptocurrency portfolios with six to twelve months volatility look-back and holding period generate statistically significant excess returns. By including a simple stop-loss rule, the downside risk of cryptocurrency portfolios is reduced markedly, and the Sharpe ratios are improved significantly.
Cryptocurrencies: Stylized Facts and Risk Based Momentum Investing#
The motivation of this research is in two folds, to understand the distributional characteristics of cryptocurrencies by means of stylized facts, and also to assess the feasibility of risk based and trend following approaches to investing in this asset class. Cryptocurrencies are more of a recent phenomenon, unlike the traditional asset classes. This poses an explicit constraint on the availability of longer history and also reliability of investment performance. Acknowledging such constraint, I focus my analysis based on the few years of data that is available.
Cryptocurrency trading: A systematic mapping study#
A comprehensive literature review mapping the academic research on cryptocurrency trading strategies. The survey covers technical analysis approaches, machine learning methods, market microstructure studies, and quantitative strategies, providing a structured overview of the state of crypto trading research.
Optimal Allocation to Cryptocurrencies in Diversified Portfolios#
This paper applies four quantitative methods for optimal allocation to Bitcoin and Ether within alternative and balanced portfolios. Using roll-forward historical simulations, all four allocation methods produce a persistent positive allocation to Bitcoin and Ether with a median allocation of about 2.7%, emphasizing the diversification benefits of cryptocurrencies as an asset class.
Risks and Returns of Cryptocurrency#
Yukun Liu and Aleh Tsyvinski (Review of Financial Studies, 2021) provide one of the foundational asset-pricing treatments of major cryptocurrencies. Studying Bitcoin, Ripple, and Ethereum, they show that crypto returns are largely distinct from stocks, currencies, and precious metals, and are better explained by crypto-native drivers such as user adoption, network activity, investor attention, and trend effects.
Our summary: this paper is the bridge between “crypto as a weird asset” and “crypto as an asset class that deserves its own factor language.” The most important takeaway is not merely that crypto is different, but that the relevant predictors are different in a systematic way. That makes this paper foundational for any crypto momentum or factor research stack: it says you should not expect stock-market factors to do most of the explanatory work.
Key metrics: the paper reports a strong time-series momentum effect in major cryptocurrencies and finds that investor-attention proxies significantly forecast future returns, while exposures to most common stock, currency, commodity, and macro factors are weak or absent.
Common Risk Factors in Cryptocurrency#
This paper develops a three-factor model for cryptocurrency returns — market, size, and momentum factors — analogous to the Fama-French model in equities. The factor model explains a substantial portion of the cross-section of crypto returns and provides a framework for evaluating whether trading strategies generate genuine alpha or simply load on common factors.
A Factor Model for Cryptocurrency Returns#
Daniele Bianchi and Mykola Babiak develop a latent-factor model for cryptocurrency returns using Instrumented Principal Component Analysis (IPCA). Rather than relying only on a small set of hand-crafted observable factors, the model extracts latent risk drivers from a large cross-section of cryptocurrency pairs while allowing factor loadings to vary with observable characteristics such as liquidity, size, reversal, and downside risk.
Our summary: this paper is the natural next step after the early crypto factor literature. Whereas Liu-Tsyvinski-Wu establish that crypto has its own factor structure, Bianchi and Babiak ask whether a more flexible latent-factor model can describe that structure better than fixed observable factors. The answer is yes: crypto returns appear to have a richer, time-varying risk architecture than simple bottom-up factor portfolios can capture.
ASRI: An Aggregated Systemic Risk Index for Cryptocurrency Markets#
This paper develops a systemic risk framework for cryptocurrency markets that simultaneously identifies high-risk periods and characterizes the market regime. The ASRI (Aggregated Systemic Risk Index) combines multiple risk indicators to provide an early warning system for periods of elevated contagion risk in crypto ecosystems.
Crypto Contagion#
This paper examines how the introduction of Bitcoin ETFs affected the diversification properties of crypto assets within traditional portfolios. The authors find that ETF introduction inverted the diversification relationship — crypto assets became more correlated with equities — demonstrating that institutional adoption changed the fundamental risk characteristics of the asset class.
The Low-Volatility Factor in Cryptocurrency Asset Selection#
Research Article #25 from Trading Research Hub. Studies the low volatility factor in cryptocurrency markets. The study examines whether selecting lower-volatility cryptocurrencies produces better risk-adjusted returns than investing in higher-volatility alternatives, as documented in traditional equity markets.
The article provides a factor-based analysis of crypto assets sorted by volatility, testing whether the low volatility premium documented in academic finance literature extends to digital assets. The practical implications for portfolio construction and asset selection in crypto are explored through systematic backtesting.
By Pedma.
Bitcoin Price Formation and Signed Order Flow#
Bitcoin prices are strongly shaped by who is aggressively buying and selling — 60% of the variation in its daily returns can be explained by signed order flow. Unlike stocks, which have multiple price anchors beyond trade imbalances, Bitcoin’s order flow explains substantially more variance. In this respect, Bitcoin resembles currencies more than equities.
The analysis, using Binance futures data, replicates prior research that found signed flow previously explained 85% of Bitcoin’s variability. While order flow’s explanatory power has decreased over time, it remains significant at 60% on average, with shorter timeframes revealing ranges from 40% to 80%.
Mentioned by Lev (@LeafiestParapet) in a discussion about Bitcoin’s flow-driven price dynamics.
By Lev.
Optimal Buy-and-Hold Asset Allocation: A Multi-Horizon Drawdown-Constrained Approach#
Yanshu Wang and Luna Wang (Peking University) investigate optimal buy-and-hold portfolio allocation using drawdown-constrained optimization across multiple investment horizons of 5, 10, 20, and 30 years. Using historical ETF data from 1996 to 2026, the paper discovers a remarkably consistent “40/40/20” framework — 40% growth equities, 40% gold, and 20% Bitcoin — that maximizes risk-adjusted returns across all tested time horizons.
Our summary: the paper’s main contribution is not the specific weights but the framework: drawdown-constrained optimization consistently lands on assets with genuinely different regime characteristics rather than piling into a single low-volatility bucket. The stability of the 40/40/20 split across horizons from 5 to 30 years is the surprising result.
Key metrics: optimized portfolios deliver Sharpe ratios 2-6x higher than conventional benchmarks. The optimal maximum drawdown constraint converges to -36% to -38% across all horizon lengths.
Mentioned by Ivan Blanco in this discussion.
Shannon’s Demon on Shitters: Volatility Harvesting on Crypto Perps#
Margin Syndicate describes an execution structure built around constant-mix rebalancing — the mechanism behind Shannon’s demon — run at derivatives leverage on netted margin across the crypto altcoin universe. The core position is constant-gross and two-legged, holding both a long and a short simultaneously and rebalancing back to a target ratio at every price threshold. Each rebalance banks the winning leg’s profit and reloads it, so the source of return is oscillation rather than direction: at leverage on margin, volatility itself becomes the revenue line. The stated philosophy is explicitly non-predictive — the same structure is run in three configurations, and which configuration applies is a reaction to the current regime, not a forecast of the next one.
The three books share one structure and differ only in the target ratio. The neutral book (50/50, zero tilt) is applied to the bulk of alts that are “printing random walks” — violent and directionless. Rather than treating that chop as untradeable noise, it is treated as inventory: a balanced book carries near-zero directional exposure and harvests the oscillation, paid by movement and agnostic to destination. The one binding condition is regime — chop pays, sustained trend charges — so these positions are audited on clip rate (rebalance frequency), not P&L, and a name that stops wiggling and starts trending is removed from the book. The signal book (tilted ratios, e.g. 60/40) is used where the desk holds an actual timeframe-based directional view. The tilt is the position (conviction sizes it), while the rebalancing wraps a volatility harvest around it and maintains the ratio through any path — with no averaging down and no scheduled adding to a losing thesis. Direction comes from the trader; maintenance comes from the machine. The depressed-asset book (modest long tilt) targets names the market has left for dead: cycle lows, crowded shorts, and negative funding that pays the long side to wait. The thesis is asymmetry — remaining downside on something already down 90% is arithmetically small while the upside on a positioning unwind is not — without predicting which specific names recover. The harvest plus the funding carry cover the cost of patience, and the whole thing is sized as one correlated basket with hard exit rules. The summary framing: three books, one structure, different ratios; hold balanced harvesters, tilt only on confirmation, and get paid to wait where the crowd is paying — and when the regime changes, the ratios change.
By Margin Syndicate.
Systematic BTC Options Report: Performance Model and Forward Projections#
A six-page report from Margin Syndicate documenting a performance model and forward Monte Carlo projection for their systematic BTC options strategy. The system executes systematic short-volatility strategies on BTC options via Deribit — short strangles, iron condors, bear call spreads, bull put spreads, jade lizards, and calendar spreads — selected dynamically from market-regime signals including IV rank, 25-delta skew, gamma exposure (GEX), and proprietary options-flow analysis. Positions are sized at roughly 1.2% capital per trade, delta-hedged via perpetual futures to maintain portfolio delta-neutrality, and run on 5-minute cycles with continuous position monitoring, assignment detection, and orphan recovery. The report is explicit that it contains a modeled track record (parameters observed from live production logs, extrapolated over a two-month window) plus forward Monte Carlo projections, and that neither section represents audited live trading returns.
The modeled parameters are disclosed up front: $50,000 initial capital on Deribit (BTC-settled), 7–21 day target DTE, adaptive IV-rank threshold, observed theta of about $167/day, current DVOL of 81.5%, a maker win rate of 81.73%, and 25-delta skew of +10.1%. Headline figures cited from the model and simulation are a Sharpe ratio of 5.83, a maximum drawdown of -1.33%, and a six-month median projected return of +13%, built from a 10,000-path simulation using fat-tail dynamics (Student-t, df=5) and volatility clustering. The report frames itself as a full methodology disclosure “built for transparency,” and repeatedly cautions that past modeled or projected performance is not indicative of future results.
By Margin Syndicate.
January 2026 Syndicate API Report: Performance Reporting Update#
A three-page investor performance report from Margin Syndicate Limited covering January 2026, when the system traded only half the month due to constrained liquidity and elevated volatility — conditions in which their “Sigma hedging” system dynamically adjusted exposure to control drawdowns. January delivered +12.3%. The report doubles as an announcement that the desk is transitioning its primary performance reporting to Quant.Space, an institutional-grade third-party analytics platform, to align with standardised, independently verifiable risk-adjusted metrics rather than the compound-return methodology used in its earlier Allo reports.
The verified metrics (Quant.Space, reporting period 27 November 2024 to 30 January 2026, ~14 months) are: cumulative return 454.0%, CAGR 329.1%, Sharpe 1.88, Sortino 2.81, Calmar 7.38, maximum drawdown -44.6%, and annualised volatility 113.1%. Trailing returns are +12.3% MTD, +32.1% three-month, and +104.8% six-month, with a best month of +44.1% and worst of -11.2%; execution quality shows a 90.22% exit-maker rate across 221,929 total trades. The narrative emphasises the Calmar ratio as the headline — strong returns without blow-up risk — and includes a methodology note explaining that historical Allo figures used (mathematically correct) compound returns for continuity, while CAGR is presented as the standardised figure for cross-strategy benchmarking.
By Margin Syndicate.
Update on the Majors: Moving Grid Capital Off BTC and ETH#
Steven Paterson (Margin Syndicate) explains a structural, regime-driven reallocation: moving grid-trading capital off BTC and ETH — not pausing it, moving it — while running overlay delta control to wind the existing grids down to completion. The rationale is not performance but market structure: BTC and ETH have started trading identically to ES and Nasdaq futures, with the same flow, operator behaviour, and construction — days of one-sided pressure into short clusters, a managed topside sweep, then distribution into the chasers. Watching all four charts on the same day, he argues, is watching one book.
That regime has a clear winner and loser. Short volatility gets paid because the whipsaw stays inside a managed range and theta collects, while grid strategies get destroyed because they need independent, two-sided price action — a preconstructed up-then-down program fills the entire ladder on the wrong side at each turn. So the allocation follows the structure: options carry the majors, grids trade alts. Alt perpetuals still have independent operators, idiosyncratic funding regimes and genuine two-sided volatility, which is where a mean-reversion system’s edge lives. The closing note is a discipline point — the system was built for mean reversion, the majors stopped offering it, and rather than argue with the tape they move the capital to where the structure still pays.
By Steven Paterson.
Systematic Market-Neutral Crypto: A Readable Statistical Signal#
Talha Çağrı Kotcioğlu shares a newly deployed market-neutral crypto strategy along with its headline metrics: Sharpe 3.45, Sortino 4.42, profit factor 1.80, and a maximum drawdown of 7.20%. The pitch is deliberately anti-black-box — no machine learning, “a statistical signal you can read top to bottom” — built and validated on two years of tick-level data with every cost baked in (taker fees, conservative slippage, and funding). It is long/short so it does not need the market to rise, uses no leverage, and is explicitly not HFT: the author says it is resistant to up to one hour of execution delay. In his backtest chart (two years, fully costed, validated out-of-sample) the equity curve shows an in-sample Sharpe of about 3.5 and out-of-sample Sharpe of about 3.3, with Sharpe by period of 3.5 (research), 3.9 (validation), 2.8 (2026) and 3.3 (clean out-of-sample), and a drawdown that peaked near 7% but stayed controlled throughout. It rode through every major liquidation cascade in the window, and the author’s stated concern is capacity, which is why he deployed it live to measure how it holds up.
How it works, per the post and the author’s replies in the comments: the strategy is built from tick-level data (rather than OHLC bars), which lets it measure order-flow and microstructure conditions finely while keeping decisions slow enough to survive a one-hour execution lag — so it can be worked into the market with a TWAP rather than racing for fills. The economic engine is a crash-convex, long-volatility profile: its edge is “pre-sizing and parity choosing”, meaning it asymmetrically tilts and sizes the book — heavily onto shorts — ahead of stress, so that liquidation cascades and violent moves are the source of PnL rather than a risk to hedge. The trade-off the author volunteers is that in flat, quiet markets the book bleeds roughly 1-2%, i.e. it pays a small carry to stay positioned for the tails. Capacity is comfortable on major coins (he notes you could TWAP over the hour and target >$10M daily volume), with the constraint coming from the smaller alt coins the portfolio also holds. The specific signal construction is not disclosed, so the “readable, no black box” claim refers to the author’s own transparency about the model rather than something a reader can reproduce from the post.
Because the post is a live-deployment announcement rather than a full methodology write-up, the comment thread is worth reading for the standard skeptic’s checklist on short crypto backtests. Several quants (Kyrylo Maltsev, Marvin Appiah-kubi) argue a two-year in-sample with a six-month out-of-sample is less than one full market cycle and likely to break under an unforeseen regime, and that the only safe use is to validate once out-of-sample and stop tuning. Mikko Ohtamaa adds the concrete failure mode that backtesting on a young venue like Hyperliquid effectively captures its 2024-2025 bull market and can fall apart afterwards. Treat the reported metrics as promising but preliminary — a live capacity experiment, not a proven multi-cycle track record.
By Talha Çağrı Kotcioğlu.
Wavelet Analysis of Cryptocurrencies — Non-Linear Dynamics in High Frequency Domains#
Kikuchi (2024) applies wavelet analysis to high-frequency cryptocurrency prices to test weak-form market efficiency, reporting cyclical persistence at different frequencies and arguing for causal structure at specific investment horizons rather than uniformly across scales.
Our summary: this is one of the very few high-frequency (rather than daily) crypto wavelet studies, which matters because the intraday bar regime is where a perpetual-futures book actually operates, and most of the wavelet-crypto literature works on daily data. The paper is read here from its abstract; it names neither the specific coins nor the sampling frequency, so the finding is directional — that predictability concentrates at particular horizons — rather than a directly reusable recipe. It complements the daily-scale evidence (Phiri, Bhuiyan et al.) that any crypto wavelet signal tends to live at shorter scales.
Algorithmic Complexity vs. Market Efficiency: Wavelet–Transformer Architectures for Cryptocurrency Forecasting#
Aldan Jay and Rafael Berlanga (Algorithms, 2026) ask whether sophisticated deep learning justifies its cost for short-term crypto forecasting. They evaluate a 2.1M-parameter wavelet-enhanced transformer that decomposes the Crypto Fear and Greed Index with a 4-level Daubechies-4 DWT into five bands (D1–D4, A4) and fuses them with twelve technical indicators, using a leakage-safe rolling decomposition (the transform is recomputed inside each 60-day input window, so no future data touches a coefficient used at time t). Benchmarking is done with Diebold–Mariano tests using HAC-corrected variance against an O(1) naive-persistence baseline.
Our summary: this is the bar any wavelet-crypto trading branch has to clear, and it is a sobering one. No model — the wavelet-transformer, ARIMA, XGBoost, LSTM, or a vanilla Transformer — significantly beats naive persistence at the 1-day horizon (DM = +19.13, p < 0.001, naive preferred). The wavelet-transformer’s RMSE is USD 2005 versus USD 1986 for naive (ratio 1.010) at roughly 3909× the inference cost, a negative return on computational investment. Ablation identifies positional encoding as the only critical component (its removal degrades RMSE by 30%); multi-seed CV is 0.77%; the naive advantage shrinks with horizon (DM 19.13 at 1 day to 6.81 at 30 days); and cross-asset transfer fails catastrophically (ETH RMSE USD 421,600, BNB USD 2.8M) because the FGI is Bitcoin-centric. The paper concludes in favour of weak-form efficiency at horizons up to 30 days, and — importantly for this collection — it is leakage-clean, which is exactly why its effect size is modest. Code and data are public.
Bitcoin and S&P 500: Co-movements of High-Order Moments in the Time-Frequency Domain#
Bouri, Kristoufek and Azoury (PLoS ONE, 2022) apply wavelet coherence and partial wavelet coherence not just to returns but to the volatility, skewness and kurtosis of Bitcoin and the S&P 500, on daily data from 2011-08-19 to 2022-01-14 (2,716 observations).
Our summary: the finding that matters for feature design is that co-movement is both moment- and frequency-dependent. The two markets are essentially disconnected before 2018, materially connected from 2018–19, with volatility linkage spiking during COVID at medium-term scales and skewness/kurtosis linkage strongest at mid-to-long scales; partial coherence conditioning on US economic policy uncertainty shows EPU intermediates the relationship. The practical implication is to build separate scale-band features per moment rather than collapsing everything into one denoised price series.
Diversification Evidence of Bitcoin and Gold from Wavelet Analysis#
Bhuiyan, Husain and Zhang (Financial Innovation, 2023) combine the continuous wavelet transform (LA(8) filter) with DCC-MGARCH on Bitcoin, gold and six equity indices, on daily data from 2014-01-07 to 2022-05-31 (2,183 observations).
Our summary: the useful, reusable output is an empirically-grounded scale banding for daily crypto features. Bitcoin shows low coherence with all six indices in the 2–16 day band across every sub-period; 16–64 day diversification is conditional and weakens in crises; and 64–256 day behaviour is mixed. That 2–16 / 16–64 / 64–256 day decomposition is a literature-backed default for how to bucket scales when engineering daily wavelet features on crypto.
Can Wavelets Produce a Clearer Picture of Weak-Form Market Efficiency in Bitcoin?#
Phiri (Eurasian Economic Review, 2022) embeds a Dickey-Fuller random-walk test into time-frequency space via the complex continuous wavelet transform, on daily Bitcoin from 2010-07-19 to 2022-03-03.
Our summary (from the abstract): predictability concentrates at higher frequencies — 2–16 day cycles — and clusters in November–February, in bubble periods, around halvings, and during COVID announcements. This is the closest thing in the crypto-wavelet corpus to a direct statement of where in scale space a signal might live, and it points at short scales rather than the low-frequency trend that most denoise-then-trade papers actually target. That directional finding is worth pairing with Bhuiyan et al.’s scale banding when deciding which bands to feed a model.
Fractal Dynamics and Wavelet Analysis: Deep Volatility and Return Properties of Bitcoin, Ethereum and Ripple#
Celeste, Corbet and Gurdgiev (Quarterly Review of Economics and Finance, 2019/2020) combine rescaled-range (Hurst exponent) analysis with wavelet analysis to study long memory and cyclical persistence across the full histories of Bitcoin, Ethereum and Ripple.
Our summary (from the abstract): the headline is that long memory and cyclical persistence/anti-persistence are found across Bitcoin’s history but not replicated for Ethereum or Ripple. That is a direct caution against assuming a BTC-fitted scale structure transfers to altcoins — which matters immediately for any cross-sectional perpetual-futures universe, and connects to the argument for channel-clustering models (WaveTS, F8) that let different coins have different scale behaviour rather than sharing one fitted decomposition.
Crypto Trend Prediction Based on Wavelet Transform and Deep Learning Algorithm#
Sumesh Eratt Parameswaran, Vidhyalavanya Ramachandran and Swati Shukla (Procedia Computer Science 235, 1179–1189, 2024; ICMLDE 2023) use the wavelet transform to split a non-stationary crypto price series into uncorrelated components, extract features from the resulting coefficients, and train a Bi-LSTM to predict the coefficients 16 days ahead. The forecast coefficients are then inverse-transformed back into a price path. SHIB is the case study, chosen for its extreme appreciation from launch, with historical data pulled from Yahoo Finance via the yfinance Python API. Accuracy is reported as RMSE against a non-wavelet baseline.
Our summary: this is a clean example of the forecast-in-coefficient-space design, and of its characteristic failure mode — which is why it is catalogued here as an instructive case rather than as a validated method. Reconstructing a price path from forecast coefficients reintroduces exactly the boundary problem documented in the leakage entries of Signal Decomposition: the inverse transform near the edge of the window draws on coefficients whose in-sample values were computed from data spanning the forecast region, so the reconstruction can look far more accurate than any causal implementation would be. The 16-day horizon compounds this, since the boundary region of a multi-level decomposition grows with the filter length and level count, and a 16-day-ahead reconstruction sits almost entirely inside it. Two further limits on how far the result travels: SHIB is selected precisely because of an extraordinary historical run, which is survivorship selection at the asset level, and RMSE on a price path is a weak proxy for whether a directional trade would have made money.
Data and reproduction: SHIB daily history from Yahoo Finance via the yfinance API, with the wavelet decomposition and Bi-LSTM described in the paper but no code release. Published open access under a Creative Commons licence, so the full text is freely available. The paper does not state the boundary-handling scheme for the wavelet transform, which is the detail that would determine whether the result survives a causal reimplementation.
Key metrics: the paper reports a lower RMSE for the wavelet-plus-Bi-LSTM scheme than for the comparison without decomposition, over a 16-day-ahead horizon on SHIB. It reports no directional accuracy, return, Sharpe ratio, drawdown or transaction-cost treatment, so there is no tradable performance figure to cite — the contribution is the architecture, not an evaluated strategy.
Scott Phillips Quant Q&A: Features, Execution and Finding Edge in Crypto#
In September 2026 Scott Phillips (@ScottPh77711570) ran an open quant question-and-answer session on X, inviting specific questions on features, edge, portfolio construction and implementation priorities, with James (@therobotjames) tagged to help field them. His framing for the session was that he would not give away anything that makes money for free, but would “close the gap” for anyone who turned up with a specific, halfway-there question — so the answers are terse, practitioner-level pointers rather than full recipes. What follows organises the substantive trading and research threads by topic; questions purely about the HyperTrend product launch are omitted.
Overfitting, robustness and knowing when you are fooling yourself. Asked how he tests for robustness given Larry Williams-style checklists, Scott plays down elaborate machinery: walk-forward is a fine idea but your most recent data is the most valuable and in crypto it is usually better to “fuckit, we ball”; separate optimisation and old out-of-sample systems are unnecessary; the honest question a robustness test really answers is “am I being a cunt?” Pressed on when a theory-backed tweak becomes overfitting he defers to friendscallmeap’s answer, and told someone data-mining directional tendencies plus statistical testing is not enough to build a reliable portfolio “unless you do hft”. A related question on the best metrics and methods to avoid overfitting on limited backtest data went unanswered.
Data quality, vendors and survivorship bias. On making sure signals are real rather than artefacts of bid-ask bounce, Scott says their signals are trustworthy mainly because they either work weakly in TradFi and pay more in crypto, or used to work in TradFi and stopped — “nothing even remotely controversial,” the book looks a lot like AQR’s. He repeatedly warns crypto data vendors are almost universally poor: the cleanest way to beat survivorship bias is to pay for point-in-time data and avoid the ballache, and paying up for curated blockchain or fundamental metrics is worth it but frustrating, and best done only after you have good trend, momentum, carry and aggression features plus solid execution.
Which features carry the edge, and what happens when they die. He runs roughly 140 features, with the contribution of seasonality features topping out at about Sharpe 3.0 from his best one — “good but not incredible.” Asked how often a retired feature’s flow mechanism truly disappears versus the price response front-running the entry, he distinguishes that from a feature merely “stopping working” — cross-sectional momentum stopped for a year and then came back, which is “totally normal.”
Signal combination and portfolio construction. On signals that look great in research but fail at the portfolio level because they cluster, Scott’s answer is that a signal is only valuable to the extent it is uncorrelated with what you already trade, and that you should build the book “like a food pyramid.” On combining features living across several clocks and horizons, he says the correct sampling method follows your priors but in practice time clocks are fine for almost everyone — you need to be very deep in crypto, or trading at ~2-second horizons, before different clocks matter.
Cross-sectional and market-neutral effects. Told that trend and momentum dominate daily-rebalanced cross-sectional books and most other effects die once you neutralise to trend, Scott’s one-word answer for the next strongest effect was “Size.” On why cross-sectional long/short performance differs between Binance and Hyperliquid, he explains that Binance has more low-quality, extractive names where low-capacity features work far better — but you cannot get size on them, so it is irrelevant for a large book and only real for the small trader.
Momentum under a hard drawdown or vol budget. Asked whether a 10% maximum-drawdown mandate rules out momentum, he says momentum is still tradable but caps you at roughly 25 vol.
Carry and funding-rate strategies. On deceleration signals layered onto a funding play, he agrees that “almost everything that works will work better that way.” For improving a volatile “dirty carry” basket he recommends common-sense rebalancing — a top-3 name slipping to 5 is probably still fine — and notes the negative-funding long leg works better hedged with BTC and HYPE. For a sub-$50k trader looking for edge he says target carry first, in the sketchiest places.
Shorting alts: baskets, squeezes, capitulation and ADL. On not getting liquidated shorting a basket of alts, the rule is to always be hedged long BTC (or BTC/ETH/HYPE), trade many different shorts, and endure the squeezes everyone knows about — that endurance is what you are paid for, which is why the trade persists. Quantifying parabolic moves and capitulation for shorting bottom-decile memecoins does help: scampump-prone coins have identifiable characteristics you can build a rolling universe around. On auto-deleveraging risk for once-a-day market-neutral strategies, he has only ever been ADL’d at prices better than he could otherwise have gotten.
Intraday momentum, order flow and venue. On prioritising local order flow versus cross-exchange confirmation and whether position-transparency data is predictive, Scott says intraday momentum improves by taking price from the discovery venue (mostly Binance) and trading where you are, while public position and liquidation signals resolve too fast for non-market-makers to profit — “mostly a nerd-knife-fight.” Related questions on maker-versus-taker edges intraday and how to choose the trading universe for intraday momentum were left unanswered.
Rebalancing frequency: daily to mid-frequency. Asked how to move from daily to hourly rebalancing without dying, the answer was to just recalculate your existing daily bars every hour — “congratulations, you are now MFT.”
Execution, market impact and building positions in illiquids. On sizing a fast (<1 minute) short-term signal, his capacity rule of thumb is never more than ~5% of ADV or 10% of open interest (and 10% is degen) — if you must get in within a minute it is probably something like liquidations. To build a big position in an illiquid alt over one to three months you stay patient and wait for a donkey to mash the button on the other side. Execution alpha at 15bps round-trip is worth chasing, but only after you have five or six decent features. And on the progression of his risk modelling and infra, he starts from “how much do I want my portfolio to wiggle” on a normal, once-a-quarter, once-a-year and once-in-several-years day, backed out into annualised vol via the square root of trading days.
Options and skew. Asked what he finds valuable in non-linear / options pricing, he says skew works much as it did in TradFi “back in the day”, with the rest reserved for DMs.
Machine learning versus economic factors. On whether ML adds edge over economic factors or just overfits faster, he says ML can work and many do it, but philosophically they prefer to transform the data carefully so the marginal benefit is small; they do not use it except in execution and portfolio construction, and it has genuinely found things mainly in HFT.
Mean reversion and regime switching. Asked for safety guidelines for mean-reversion strategies and thoughts on regime switching, his answer ties safety to caps on portfolio percentage per coin and position size versus ADV, tested in the real world rather than “mathland,” and grouping similar names — while flatly stating regime switching “doesn’t work.”
Where a small trader should hunt: sketchy venues, memecoins and arb. For very low-capacity edge tradable with a four-figure stack, he points to small coins listed only on tiny fourth-tier exchanges, reading the product docs carefully, and either farming funding or manipulating spot via a long perp plus a max-slippage on-chain buy. On overlooked exchanges the principle is that the sketchier the venue the more likely you can make money — and the less likely you can withdraw. Cross-chain DEX arb is lower priority than CEX-DEX arb. On edge in Robinhood-chain memecoins and wallet tracking he notes memecoins with a multisig are more professional and pump-prone, and to think in terms of the characteristics of winning teams and coins; on trading memecoin volatility off abnormal volume he explains that volume is autocorrelated — rising volume and number-of-trades are “pumpy.”
The recurring answer: community is the real alpha. Across several replies Scott insists the highest-return activity for a developing trader is getting into a group chat with people who make money and making yourself either deeply useful or funny — stated to the small trader looking for edge and to the memecoin hunter asking how to get into those networks.
Learning to run execution. Asked what he had to learn building out execution and managing people, he answers candidly that it is really, really hard, that he “didn’t know shit about fuck,” and that his friends are very smart.
By Scott Phillips (@ScottPh77711570), with James (@therobotjames).
People involved in the discussion: @ScottPh77711570, @therobotjames, @Avik_Saikat, @Cyespx, @fin_constructor, @wiltonhotz, @dima_quant, @quant_arb, @HF_Trader, @trad62866, @HerrOak, @AnyonCurve, @risk_taker88, @JstSomeGuy, @olvehe, @PostTradeZen, @SidKZ0, @eyshu_tw, @AffectiveAutism, @jxy610, @6enginecig, @gotRekt1337, @OnlyDeFiGuy, @Squankulous, @yungp9rple, @299792458MaxHH, @NikolasBohm, @b_ruug, @simnmayer, @123skely, @_unyk, @BalefulWatcher, @Vladimi72981101, @lifeenjoyer___, @Metinique1, @derivative_bro, @miquel_malaga, @RobinLi247502, @vndeniable, @markovschainsaw, @DMTCapital, @0xelevenquit, @HazeGenesis, @wock9000, @DPhantomOfDPits, @quant_spence, @meurfault, @kwertytwo, @Swan_Pineapple, @warp_sys, @etant0x, @idiotek77777, @predictsnfn.
Bitcoin Beta De-noising: Isolating Altcoin-Specific Trading Signals#
Research Article #34 from Trading Research Hub. Presents a step-by-step guide to isolating asset-specific signals by removing Bitcoin’s dominant influence from altcoin price movements. The technique of “beta de-noising” separates the idiosyncratic component of an altcoin’s returns from its correlation with Bitcoin.
The article demonstrates how this de-noised signal can improve predictive accuracy for trading strategies on assets like Ethereum. By stripping away the market-wide factor (Bitcoin’s influence), traders can identify cleaner asset-specific signals that may provide genuine alpha rather than simply proxying for Bitcoin momentum.
By Pedma.
Cross-Sectional Alpha Factors in Crypto: 2+ Sharpe Ratio Without Overfitting#
An in-depth examination of how traditional quantitative finance factor models — originally developed for equities by Fama and French — can be adapted to cryptocurrency markets to generate alpha. The authors argue these factors still qualify as alpha in crypto because the market hasn’t fully arbitraged them away. The methodology involves constructing long/short factor portfolios across the top 50 digital assets by market cap: ranking tokens daily by a chosen metric, going long the top 20% and short the bottom 20% to isolate factor exposure while maintaining zero directional market exposure.
The research focuses on two primary factors: cross-sectional momentum (which operates on ~30-day horizons in crypto, much shorter than the 6–12 months typical in equities) and carry (exploiting perpetual futures funding rate differentials using open-interest-weighted composite rates across exchanges). By combining momentum and carry into a multi-factor portfolio, the researchers achieve a 2+ Sharpe ratio. Key practical challenges addressed include survivorship bias from delisted tokens, extreme volatility dispersion between Bitcoin and memecoins, and limited reliable data before 2020. The authors apply inverse volatility weighting and volume filters to manage these issues, and find that diversification benefits plateau at roughly six factors with achievable Sharpe ratios around 2.5 for single portfolios.
By Mark Aron Szulyovszky and Daniel Szemerey.
How Speculative Money Flows into Crypto#
A framework for understanding and predicting speculative capital flows into cryptocurrency markets, structured across three concentric layers. The outermost layer examines macro/global factors including momentum spillover effects from other asset classes — particularly cloud computing sector momentum as a leading indicator for Bitcoin — and Bitcoin-Gold correlation as a proxy for institutional safe-haven perception. The middle layer tracks ecosystem entry points via stablecoin market capitalization, stablecoin exchange inflows (capital “parked for purchasing”), and Bitcoin spot ETF netflows representing institutional interest.
The innermost layer monitors internal leverage metrics: perpetual futures funding rates, liquidation volumes, long/short position ratios, and margin borrowing rates. The author recommends reducing directional exposure when speculative flows decelerate and proactively cutting positions during over-leveraged periods before liquidation cascades. These signals prove even more effective on smaller-cap, more speculative assets like Ethereum. The key insight is that traditional volatility-based risk management is insufficient given crypto’s leverage concentration and lack of circuit-breakers — incorporating exogenous money-flow and liquidity metrics provides a superior risk framework for digital asset portfolio management.
By Mark Aron Szulyovszky.