28% annualised. That's the headline result of a Springer academic implementation that back-tested a multi-factor quantamental model on Chinese equities. The model blended accounting ratios such as book-to-price and return on assets with statistical regression and machine learning, producing large historical excess returns in sample. Institutional managers, active fundamental investors and sophisticated retail traders are already using these hybrid workflows, but researchers say the practical next step is open replication and out-of-sample testing.

The central claim is simple and precise: combine sound accounting signals with quantitative tools and you can produce historically large returns. The Springer chapter reported a 28 percent annualised return in its Chinese-market back test, a striking figure that explains why quantamental approaches have moved from academic curiosity to mainstream practice among institutional and retail players.

What the factors are

Factor research groups signals into broad buckets, and four of the categories most tightly linked to accounting data are Value, Size, High Yield and Quality. MSCI groups these categories in the same way. Value measures cheapness relative to fundamentals using book-to-price, earnings-to-price and cash-flow ratios. Size looks at market capitalisation and free-float capital. High Yield captures dividend income via dividend yield. Quality measures financial health through return on equity, return on assets, leverage and earnings consistency.

A recent primer published on Medium that synthesised academic evidence highlights two signals that repeatedly stand out across US studies: the enterprise multiple and return on assets. That primer points to t-statistics as the clearest yardstick, with a t-stat above 2.0 taken as evidence of significance and above 3.0 as very strong. Those thresholds are useful because they separate noisy back-test discoveries from persistent effects worth trading.

Quantamental implementation sits on three operational levels, a framework Interactive Brokers lays out. First, asset-level work augments a single-stock fundamental model with alternative data and machine learning. Second, event-driven approaches use quantitative signals to trade around earnings, macro releases or industry shocks. Third, portfolio-level construction blends factor tilts with risk-parity or optimisation constraints so that a multi-factor view survives real-world limits like liquidity and concentration rules.

Mastheads

Practitioners such as Interactive Brokers and QuantInsti describe the building blocks in practical terms. Routine quantamental tools include automated screening on conventional ratios such as price-to-earnings and price-to-book, enterprise-value-to-sales, and debt metrics. Natural-language processing is applied to filings to extract signals from management commentary.

And alternative datasets convert unstructured information into tradable inputs: foot-traffic analytics to predict store sales or credit-card expenditure series to proxy consumer spending are both cited examples used to feed factor models.

Those techniques help solve the second source of edge the literature describes. The first source is economic logic: a firm with durable accounting strength should, all else equal, generate higher expected cash flows, which supports exposure to value and quality signals. The second is scalability: quantitative tools let teams systematically screen, normalise and combine those signals across thousands of firms, and apply portfolio construction rules that a single analyst cannot.

The Open Asset Pricing project, presented by Chen and Zimmermann in 2022, is an example of the replication infrastructure the field needs. It's an open repository that replicates thousands of published factors and lets researchers test whether signals survive out-of-sample scrutiny. That kind of replication is exactly the safeguard the Medium primer recommends against the factor zoo problem, where many published effects fail to hold up after discovery.

Empirical reliability is the recurring caution in both academic and practitioner accounts. The Springer chapter that recorded the 28 percent annualised back test also points out methodological caveats.

Back-tested returns can overstate what a live strategy will deliver if researchers don't control for data snooping, sample selection and look-ahead bias. In plain terms, a model that fits historical quirks may not survive the next market cycle unless it's validated on fresh data and constructed with realistic trading assumptions.

In practice that means two things. First, screen candidate factors in broad replication datasets such as the Open Asset Pricing repository to check whether they were rediscoveries or genuinely new effects. Second, run out-of-sample tests and realistic simulations that include transaction costs and position limits before committing capital. Those steps are the difference between an interesting back test and an investable strategy.

Related Articles

Researchers and practitioners point to a clear next step: validate candidate factors in open replication datasets such as Open Asset Pricing and run out-of-sample tests before committing capital. Originally reported by interactivebrokers.com.

This article was created with AI assistance.