It’s a big day for us: we’re opening Aperiodic and our whole data catalog for the public.

It’s been long time in the making. We’ve been customers to all crypto data platforms and came out disappointed with the quality and variety of offering. For anyone who’s remotely serious trying to build quantitative strategies, you need:

• Point-in-time data

• Basic and advanced liquidity metrics (slippage, bid/ask price twap/vwap, top-of-the-book and deep aggregate order book liquidity)

• A large variety of unique market microstructure metrics that describe the state of the market well - (Run imbalance, small / larger order flow volume,

• Full exchange coverage without survivorship bias

• Local-timestamped versions of all time series

Subscribe now

You may get the maximum of 1 or 2 of these with the existing providers - and all with Aperiodic.

What’s in it?

Over the last couple of years, we’ve run hundreds of queries on L1, L2 data to test hypotheses around what kind of metrics could have alpha. We’re opening up the whole market microstructure catalog to the public.

Unlike others, Aperiodic gives you full history for your subscription price.

A couple of unique metrics available that’s not offered elsewhere:

Deep L2 Imbalances — Bid/ask quantity imbalance aggregated across multiple order book depths (5, 10, 20, 25 levels), rather than just the top of book.

Flow Entropy — A Shannon entropy measure applied to the distribution of buy and sell trade arrivals over a window, quantifying how “disordered” or unpredictable the order flow is. Low entropy signals a structured, directional market where trend-following strategies tend to be more effective, while high entropy reflects choppy, mean-reverting conditions where directional signals break down.

Slippage — The realized cost of executing a market order against the live L1 order book.

Kyle lambda — An empirical estimate of Kyle’s λ, measuring price impact per unit of signed order flow.

Large trade reversal — A metric capturing how much of the price move driven by a large aggressive trade subsequently reverts within the interval — distinguishing informed flow (which produces permanent impact) from uninformed or liquidity-driven flow (which mean-reverts).

Browse more than 100+ metrics over here!

The most convenient data interface we could imagine

Aperiodic is designed for the you: the API returns pre-signed URLs of ready-to-use parquet files that you can download with arbitrary concurrency (no rate limits), instead of an arbitrary format that you’d need to parse & figure out how to store.

We released a lightweight python wrapper here as well.

Download all historical data for a set of metrics for all symbols in a couple of lines of code.

We also offer direct Object Store bucket access for yearly Prime subscribers.

Data infrastructure you don’t have to build

Over the last couple of years, we’ve ingested hundreds of Terabytes of crypto market data, and built a very effective infrastructure that lets us query it in any way imaginable.

The only alternative up until today to access these metrics are:

1. buy raw tick / L2 data (very expensive)

2. figure out a way to store it (could be even more expensive)

3. query it (expensive, weeks of work)

4. keep it up-to-date (months of work)

Paying tens of thousands of dollars in the process to cloud providers and for a small team of experienced data engineers.

Aperiodic will get you access to these metrics so you can focus on your research & analysis from day 1, instead of spending months on data engineering pipelines.

What would you like to see next?

We maintain an collaborative, ambitious roadmap and very much welcome your feedback on the most useful additions: Aperiodic Roadmap.

Please shoot us a message if you have any ideas (metrics, delivery methods, etc.) you’d like to see here.

We hope Aperiodic expands your access to data that was previously out of reach; to celebrate the launch, we’re offering 40% off annual subscriptions!

Subscribe now

One of our biggest factor research breakthrough came when we made turnover a first class citizen. The lesson was: high IC is essential, but not sufficient by itself.

Subscribe now

It’s easy to skim over the fact that a high turnover factor is not directly comparable to a low turnover one - they’re qualitatively different, and will exhibit fundamentally more or less (statistical) significance.

We also started to systematically turn unprofitable, too-high turnover factors into profitable ones by applying simple smoothing / moving averages.

Example: Altair, our highest variance factor, flips the portfolio almost daily—completely unrealistic once costs are included. To keep ourselves honest and avoid overfitting, we target a daily turnover of 30-40% for any factor (that has a higher turnover by default) — before we even look at performance.

This avoids the usual bias towards performance, as well as it makes the factors comparable.

See the difference yourself: “Altair” raw (with 100% daily turnover)

Interactive version

When we apply a 20-day moving average:

Interactive version

As always, happy to connect on LinkedIn and talk shop!

Catch you in a couple of days with another short post!

In the early ’90s, the quant forefathers (Fama and French) introduced their now-canonical factor models: first three, then five, and eventually seven, explaining much of the variation in US equity returns. Today, these models are used to understand what easy-to-replicate risk factors managers are being exposed to. Allocators will want to pay for only the truly unique return sources that can’t be replicated with a couple of lines of SQL.

Crypto markets are currently in the state where many of these “traditional” factors can be still considered alpha, with considerable return potential.

Subscribe now

It’s important to note that these cross-sectional factors work without any directional exposure - the regular crypto boom-bust cycle may affect how effective they are, but by design, they have 50-50% long and short exposure at any point in time.

How a Long/Short Factor Portfolio Works — in Brief

1. Define the universe – e.g., the top 50 digital assets by market cap, updated on a rolling basis to avoid any lookahead bias.

2. Choose a factor metric – for example, momentum measured as the percentage return over the past n days.

3. Rank the assets daily based on that metric.

4. Go long the leaders, short the laggards – take equal-weighted long positions in the top 20% and equal-weighted short positions in the bottom 20%.

This structure isolates the factor exposure while remaining market-neutral.

Cross-sectional Momentum

Cross-sectional momentum is a well-documented anomaly: assets that outperform over a given formation period (“winners”) tend to continue outperforming those that underperform (“losers”)—at least in equities.

The same applies to digital assets, with a couple of caveats:

• Shorter horizons – In equities, momentum signals often span 6–12 months; in crypto, they’re concentrated over much shorter windows (~30-day formation periods) and decay rapidly, sometimes reversing within months.

• Crash-prone behavior – Momentum crashes in crypto are typically triggered by violent reversals in a handful of assets (e.g., short-squeezed losers), rather than broad market shifts.

• Large-cap dominance – Most of the returns come from top-40 market-cap coins; smaller tokens suffer from liquidity constraints and higher idiosyncratic risk.

• Regime sensitivity – Performance weakens in low-sentiment market regimes.

There’s clear evidence that even the simplest long-only, time-series momentum strategy on Bitcoin have consistently outperforms in risk-adjusted terms - but it works well in the cross-section as well.

Carry in the cross-section

Subscribe now

Unlike momentum, the carry factor traces its roots to commodity markets, where futures often trade at sustained premiums or discounts to spot prices due to factors like storage costs, seasonal patterns, and supply constraints. By expiry, these gaps must close, creating a well-defined arbitrage opportunity.

In crypto, carry is most easily measured via perpetual futures funding rates—periodic payments between traders designed to keep perpetual prices anchored to spot. When perpetuals trade above spot, longs pay shorts; when they trade below, shorts pay longs.

There are two ways to implement carry in this context:

• Hard arbitrage – Short assets with high funding rates, buy the same notional exposure in spot, and hedge out any price risk. Profits come directly from the funding payments.

• Soft (cross-sectional) carry – Short the assets with the highest funding rates and go long those with the lowest (or most negative), without explicit hedging. Here, returns come from both the funding spread and differences in expected asset performance.

Funding rates vary across exchanges, so we use an open-interest-weighted composite to avoid exchange-specific distortions and better capture the underlying carry signal.

The Caveats

Building crypto-specific, market-neutral, cross-sectional portfolios is far from straightforward. In equities, short availability is a primary constraint; in crypto, the bigger hurdle is survivorship bias—it’s often difficult to source reliable data for delisted or inactive tokens.

Beyond that, most crypto assets—outside of a handful of large-cap names—behave more like penny stocks than mid-cap companies: thinly traded, highly volatile, and frequently subject to pump-and-dump dynamics. Historical testing before ~2020 is of limited use, as the market only reached a comparable level of maturity in the past five years.

Even in this recent period, creating a stable, tradable universe is challenging. We use market cap ranking with additional volume, volatility, and availability filters, with smoothing to limit turnover and avoid overreacting to temporary shifts in liquidity (memecoin spikes).

Another key complication is the dispersion of volatility within the tradable universe. In equities, volatility differences are significant; in crypto, they’re extreme - Bitcoin and the average memecoin might as well be different asset classes. Equal notional long and short positions can still leave the portfolio heavily tilted toward the more volatile side. To address this we apply inverse volatility weighting at the portfolio level so highly volatile tokens get proportionally smaller allocations.

But what’s the alternative to cross-sectional strategies? Potentially overfitting specific signals to individual assets, or abandoning market neutrality entirely.

Achieving Sharpe Ratio of 2+ without Overfitting

No single factor can consistently forecast returns across all market regimes—it’s an unrealistic expectation. Probably everyone is tired of the mantra “diversification is the only free lunch” - but diversification works particularly well in the context of multi-factor portfolios. The combination doesn’t need to be particularly sophisticated either, even a simple arithmetic average of signals can deliver tangible benefits:

• Reduced turnover – When factors disagree, offsetting positions naturally cancel out.

• Greatly improved risk-adjusted returns – Uncorrelated return streams smooth the equity curve and raise Sharpe.

When we put these 2 (Momentum + Enhanced Carry) established, well-known factors in a single, portfolio, we get something we call “Foundational” portfolio:

How Far Can This Scale?

Diversification may be the only free lunch, but even free lunches have limits. Once you ensemble more than about six factors, the marginal benefit drops to near zero. We currently track ~20 alpha factors (50 overall) and have found it’s easier to build three orthogonal portfolios with Sharpe ≈ 2.5 than to squeeze a single portfolio up to Sharpe 3.

This breadth puts us in a unique position: we can selectively publish and license some of our proprietary factors without compromising our core edge. You can see their live performance on Unravel, and if you’d like to dig deeper, reach out.

As always, happy to connect on LinkedIn and talk shop!

* The Sharpe Ratio of “Foundational” is actually less than 2, but we track other multi-factor portfolios with considerably higher risk-adjusted returns.

We’ve outlined a base framework to systematically identify and measure the drivers of crypto liquidity—beginning with external macro factors and progressively focusing inward on ecosystem-specific effects. With an emphasis on the qualitative aspects of these factors — specifically, measuring and reporting on their predictive power.

1. Outside the ecosystem: Global / Macro Factors

These are usually slow moving factors — the kind that talking heads on CNBC would be talking about, if they were talking about crypto:

Speculative Flow - The overall liquidity that’s moving towards more speculative, risky assets, like: tech stocks, emerging markets, high yield bonds.

These has been the quantitatively most consistently predictive factors over the last 5 years. Technically, they should be called “Momentum Spillover” effects - the momentum of other asset classes heavily affect the crypto market, creating predictable patterns that can be effectively used to manage risk.

For example, the momentum of cloud computing stocks has been leading the overall crypto market’s returns.

Safe heaven perception - When the current perception of Bitcoin is closer to Gold than a speculative asset, it’ll behave differently during times of stress and “risk-off” events. We could quantify this relationship by measuring the rolling correlation coefficient between Bitcoin and Gold’s returns, with the following results:

We see that the higher Bitcoin’s correlation was with gold, the higher its forward returns are. And, that the opposite this is true when measuring Bitcoin’s correlation with tech equities.

Subscribe now

2. Closer to the Ecosystem: Money entering the sidelines

Stablecoin Market Cap (Indirect) - Tether and other stablecoin providers report their tokens in circulation. But as of lately, there’s increased adoption in stablecoins to settle native currency transactions - and not all of it is going to crypto. There are more fine-grained measures of that may be more reliable, see below.

Stablecoin Exchange Inflow & Netflow (More direct) - Exchanges publish their wallet addresses for transparency reasons — we track the largest 40 exchanges stablecoin in & outflows, which enables a closer look at what’s happening within the stablecoin ecosystem.

The rationale is: there are very few reasons to park stablecoins on exchanges other than considering buying something with them.

ETF Netflow (Direct) - The spot Bitcoin ETF was launched in January 2024, had quite a good run since then, and it’s commonly understood as a proxy for institutional interest. There’s transparency here as well, providers publish the amount that entities holds, daily. Thanks to market makers, the ETFs are are - during normal circumstances - keep their NAV closely aligned with their market price, and that money flows into crypto effectively immediately through them — making this more of a direct channel for money to enter the crypto markets.

And evaluating its predictive power is tricky, as there’s only a bit more than a year of history available - which also coincided with a major bull run, making it very hard to judge whether it represents skill, we’ll dig into it in a later article.

3. Inside the Ecosystem: Leverage

Leverage is an effective indication of crypto system-wide risk, as it’s increasing the likelihood of sudden de-leveraging effects & large amount of forced liquidations.

This is exacerbated by the lack of circuit-breakers — a safety net that equity markets usually possess — and the result is usually a 10-20% wipeout of the whole ecosystem’s market cap within an hour.

Estimating leverage in the ecosystem can be done with various metrics, including:

Funding rates of perpetual futures - the “carry” that traders need to pay in order to maintain their position on (leveraged) futures contracts. The higher, the more people are willing to pay a premium to stay long. The only reason for that is to maintain leverage, making it a good candidate for a leverage proxy metric.

Liquidations - a large round of liquidations (caused by large moves) usually means leverage is flushed out of the system: you can only get liquidated by being over-leveraged.

Long / short ratio - the volume of long positions, divided by the shorts. At the extremes, it can highlight the fact that the vast majority of the participants are already “very long” and the market can’t be really stretched further.

Margin borrow rates - traders usually borrow stablecoins to long other assets. The higher this rate is, the more people are borrowing, making it a relevant proxy metric.

We’ll explore these factors in an upcoming article, keep tuned!

Subscribe now

Aside from the ecosystem-wide factors, there are also asset-specific ones, mainly along the lines of: public perception (trends/ fads), fundamentals (usefulness), and sentiment & momentum, which are usually difficult to disentangle.

How to use these factors?

The best framing for these factors is: they’re risk proxies, covering valuable, otherwise ignored aspects of the market, and can be used effectively to size directional exposure.

Traditionally, if you wanted to manage your portfolio’s risk with quantitative methods, you’d be relying on recent volatility to cut size when markets are volatile. But, it works less well in crypto — we’ll cover this later. It’s not that it doesn’t work because many market participants are already using this technique, more that the large amount of leverage and the sudden liquidity shocks mimic market-wide volatility-based risk management and erode its usefulness.

What would an effective risk management layer look like, that also includes exogenous risk factors?

1. When less-to-no money is flowing into the crypto ecosystem, reduce directional exposure. Especially when the metrics indicate that money is about to start flowing out of crypto, like people moving their crypto assets to exchanges.

2. When ecosystem-wide leverage is the highest, the risk of an even smaller shock creating an cascading sell-off is higher. Proactively reducing exposure during prolonged over-leveraged periods is beneficial - before an inevitable, sudden de-leveraging event happens. By that time, it’s already too late: liquidity dries up and there’s no effective way to hedge anymore.

These risk factors work even better when applied on the more speculative assets, starting from Ethereum and going down in market cap — because their returns are more purely driven by speculative money flow & leverage.

We realized the value of broadening risk metrics beyond traditional measures by incorporating money flow and liquidity indicators— with Unravel, we’re working on making this research accessible and easy to apply.

Especially for risk management is needed the most: for the most volatile assets.

As always, happy to connect on LinkedIn or X!

Subscribe now

Volatility targeting’s goal is to keep an asset or portfolio’s volatility within reasonable boundaries, by actively adjusting exposure - it’s among the very few effective, universal techniques in quantitative finance. I can’t really think of many other than “diversification” and volatility targeting, if I had to answer on the spot.

A quick intro: inverse volatility scaling means when an asset is getting more volatile, you should reduce your exposure to it. People usually measure volatility with a rolling standard deviation of an asset’s returns. It looks usually something like this:

Volatility usually clusters — when the markets are volatile, they stay volatile for some time. You can kind of see it on the chart above, it doesn’t go “go back immediately” to its starting position1. But, it does mean revert after some time.

Volatility targeting is using inverse volatility scaling with a pre-defined (annualized) volatility target level. The S&P500’s annualized volatility is around 16%, Bitcoin’s is around 60%.

You can make your strategy, or asset’s cumulative returns comparable by using the same volatility target on them. It’s also a good way to normalize the risk contribution of each asset within a portfolio - so a really volatile asset won’t dominate it.

What’s the opposite of volatility targeting?

I’d say “buying the dip” is a pretty good contender - as inverse volatility scaling can be translated into: “reducing your exposure on a dip”.

And buying the dip is also effective. I think fundamentally almost all systematic strategies that trade the SP500 only a couple times per year, long-only, are quantitative equivalent of “buying-the-dip” (and ideally should be called as such, for transparency reasons). Retail investors kind of intuitively understand its effectiveness.

If you plot the average 30 day forward returns when the SP500 was in a drawdown, you’d see this2:

This chart indicates that you should have bought the most when SP500 is in the biggest drawdown — thank you for that, captain hindsight! Of course this is not something that can be effectively executed, and definitely can not be called a risk management technique.

There are fundamental differences here that makes this comparison impossible: buy-the-dip assumes you have cash to allocate, while volatility targeting assumes that you are fully invested (calculated based on the current volatility level).

Although you could add to a position if it gets volatile (which usually means its price is falling), that’d would be more of “buy more when in drawdown” technique. But that sounds pretty dangerous, kind of the opposite of a risk management strategy.

Subscribe now

The paradox

We looked at what the average returns were when the SP500 was in different drawdowns. Let’s look at what they are when volatility is in a certain range3:

When volatility was really high, we get higher average forward returns, than in the low volatility regime - there’s a catch of course. But that’s the opposite of what we’re looking for!

Let’s look at the simplest possible backtest we could create, where we use volatility targeting with a 20 day rolling window on the SP500, rebalancing daily, with fixed 0.1% transaction costs:

You usually get 10-20% increase in risk-adjusted returns when you “blindly”4 apply volatility targeting on an asset.5

How is this possible? How come

It’s mainly because of “volatility drag” - the difference between cumulative average returns (a multiplicative process) and average returns (an additive process). We could mitigate its effect by averaging log returns instead of arithmetic returns, per Kris’ explanation.

Try it yourself

We released a small python package (with MIT Licence) that helps you apply volatility targeting on any return series.

It’s a collection of transparent, easy-to-audit functions, that we wrapped it into a package called risklab on pypi — it’s super easy to use:

vol_signal = scale_to_target_volatility(
    target_volatility=0.16, # the annualized volatility target
    rolling_window=20,
    returns=sp500_data["Close"].pct_change(),
    upper_limit=2.0,
    lag=0,
    fill_initial_period_with_mean=False,
    annualization_period=252,
)
results = backtest_signal(
    signal=vol_signal,
    underlying=sp500_data["Close"].pct_change(),
    transaction_cost=0.001,
    lag=0,
)

Explore what asset classes volatility targeting works best for using risklab and this notebook.

Subscribe now

Is this a silver bullet? What am I missing?

I personally always felt a bit cautious using only past volatility to manage risk. It basically means you’re over-leveraged at times when volatility is very low, to compensate for reduced leverage when its high. Looks fine in a backtest, but it’s a lot more uncomfortable when you are actually trading.

The missing piece is the realization that:

not all periods of very low volatility are created equal.

Unexpected events may hit when:

• Volume, liquidity is structurally low

• Implied volatility and the volatility risk premium is low

• Leverage is high

In those cases, you probably see massive moves, “de-leveraging” events where people have to reduce their exposure. Some may call it “Risk parity sell-of” or “Carry trade unwinds” - and their effect are probably exaggerated by the widespread usage of volatility-based risk management itself!

Last summer, implied & realized volatility were all at the lowest points in the last couple of years. VIX term structure was extremely flat, volumes were low (usual summer break vibes) — all of these were signs that a small sell-off could trigger a lot bigger one.

Can we use these and similar, asset-specific, or market-wide risk factors to reduce the harm of the sudden de-leveraging events, like August 2024? I believe so — you can explore them for crypto here, and will cover more assets soon.

We work on Unravel, where we make forward-looking, exogenous risk factors available and easy-to-use - starting with the most volatile assets.

As always, happy to connect on LinkedIn or X.

Subscribe now

This is part 1 of an in-depth investigation of how quantitative risk management could help improve risk-adjusted returns. We’ll cover the mechanics, the implementation details, benchmark the different “risk proxies”, ensembling and a lot more!