Quantitative Research and Consulting

The difference between a portfolio that generates consistent alpha and one that merely tracks factor exposure often comes down to the quality and proprietary nature of the underlying research. Commercially available factors, accessible to every quantitative shop running the same data feeds, converge to lower alpha over time as more capital follows the same signals according to academic research by McLean & Pontiff. Building durable investment edge requires proprietary research infrastructure: custom factors derived from data that is not widely held, models built to capture non-linear relationships, and a continuous improvement process that keeps the signal ahead of the market.

The Stance Capital team has been building this kind of infrastructure since 2013. This page describes our quantitative research and consulting capabilities: what we build, how we build it, and how we partner with institutions, advisors, and asset managers to develop proprietary investment models over multi-year engagements. Advisors and institutions who want to deploy existing Stance strategies through a platform rather than build proprietary models should see the Model Delivery page.

Who This Is For

Registered investment advisors building proprietary model portfolios need quantitative differentiation to justify active management fees. A proprietary factor model, particularly one incorporating alternative data signals not available to competitors, creates defensible differentiation without requiring a full internal research team.

Institutional asset managers running systematic or quantitatively-informed strategies can augment their existing research infrastructure with proprietary factors developed on a collaborative basis. This is particularly relevant for managers who currently rely on commercial factor libraries and want to move toward genuinely proprietary signals.

Foundations and endowments benefit from sustainable and governance-integrated quantitative strategies designed to manage tail risk and identify high-quality compounders over multi-decade horizons. Institutions with explicit sustainability or mission-aligned mandates benefit from factor integration rather than static exclusion screens.

Family offices with significant equity portfolios often have mandates flexible enough to deploy quantitative strategies not available through the traditional fund universe. A custom factor model built to specific risk preferences, factor tilts, and sustainability requirements provides an investment process tailored to a single client.

Stance's Quantitative Capabilities

Proprietary Factor Library

Stance's proprietary factor library includes signals derived from both standard and alternative data sources.

Using standard data: Options-derived factors; relative value estimates; time series forecasts; custom momentum models; relative value model.

Using alternative data: Flow forecasts; insider transaction models; ownership models; sentiment models; sales growth forecasting across one, two, and three-plus quarter horizons.

The Custom Portfolio Optimizer

Most quantitative programs rely on third-party optimization tools. These are off-the-shelf solutions that, like off-the-shelf factors, do not generate edge on their own. Stance has developed a proprietary portfolio optimizer designed to implement factor immunization approaches and translate model forecasts into portfolio positioning with fine-grained control over risk, turnover, and factor exposure. The optimizer is core research infrastructure, not a third-party plug-in.

Machine Learning Infrastructure

Our ML infrastructure supports ensemble models built across multiple algorithms and time horizons, with automated model monitoring, statistical validation, and full pipeline management from raw data ingestion through final forecast generation. The pipeline is designed to be automated end-to-end, reducing operational risk and enabling consistent, scalable model execution.

Published Research and Open-Source Contributions

Stance team members have published papers and delivered conference presentations on the use of alternative data for factor construction and portfolio optimization. We have contributed to open-source projects in the quantitative finance space. This engagement with the research community keeps us connected to the frontier of the field.

The Factor Spectrum

The starting point for any quantitative investment program is where your signals come from. The answer determines whether your program has a structural advantage or is simply paying for market access.

Off-the-Shelf Factors

Commercially available factor libraries offer hundreds of pre-built signals spanning value, momentum, quality, low volatility, and growth. These factors are inexpensive, well-documented, and widely used. They are also commoditized. When the same factor is available to every manager running the same data provider, the primary available source of differentiation is factor timing, which is a weak and inconsistent approach according to academic research by Asness, Ilmanen, & Maloney. Off-the-shelf factors produce off-the-shelf results according to academic research by McLean & Pontiff.

Proprietary Factors with Standard Data

The next tier involves deriving custom signals from standard, widely available data, including earnings reports, price and volume, and analyst estimates, by applying proprietary methodologies to extract forecasts not available off the shelf. The resulting signals are not directly replicable even when the underlying data sources are known, because the methodology itself is the IP.

Proprietary Factors with Alternative Data

The strongest durable differentiation comes from factors derived from non-traditional data sources not widely available in the market. Alternative data, from options flow, insider transactions, satellite imagery, proprietary surveys, and similar sources, creates signals that are difficult for competitors to replicate. When multiple alternative data vendors are blended, the resulting factor becomes more defensible: it cannot be reproduced by a competitor subscribing to any single source. The further along this spectrum your program sits, the more valuable the institutional IP you are building.

Alternative Data

The Benefits

The primary benefit of alternative data factors is proprietary institutional IP. Non-traditional data helps ensure that derived factors remain proprietary, because the data itself is not widely available or requires significant processing capability to use effectively. When multiple vendors are blended, combining signals from two or more non-overlapping alternative data sources, the resulting factor cannot be reproduced by a competitor subscribing to any individual source.

The Challenges and Why They Are Manageable

Alternative data comes with real operational challenges: data quality is often lower than traditional financial data; procurement can be difficult and expensive; processing raw data into usable features requires significant engineering; and connecting multiple vendors into a consistent pipeline adds technical complexity. Feature engineering, the process of constructing the right model inputs from raw data, is often where the most value is created and where the most depth of experience is required. These challenges are manageable with the right infrastructure. The operational overhead of working with alternative data is itself a barrier to entry: it filters out less-resourced competitors and rewards those who invest in building the capability properly.

Machine Learning

Machine learning, as applied to quantitative investing, is the use of statistical learning algorithms to identify patterns in data that generate forecasts of security returns. This is different from generative AI, which produces text or code, and is more closely aligned with what quantitative practitioners have done for decades, augmented by modern computational methods. The Stance Capital team has been deploying ML for alpha models since 2013.

Attribution Models vs. Alpha Models

Not all factor models generate alpha. Attribution models are designed to explain what happened: to decompose past performance into factor contributions. These are useful for risk management and reporting, but they are backward-looking by construction. What best explains the past is not necessarily a useful predictor of the future.

Alpha models are designed to forecast returns, to identify which securities are likely to outperform or underperform over a forward horizon. Strong alpha factors are generators of predictive signal that persist over time. They may not cleanly explain past performance the way attribution models do. The objective is forecasting, not explanation. Stance builds alpha models.

Ensemble Methods: Why One Model Is Never Enough

A single learning algorithm trained on a single history of data will be overfit to some regime or feature set of the market. Ensemble learning addresses this by combining multiple learning algorithms and variable selection approaches into a blended forecast that captures different types of relationships in the data. Models are applied across different time horizons and market conditions, with outputs weighted and combined into a final forecast. This process is automated and continuously monitored. New modelling techniques can be evaluated and added to the ensemble if they pass statistical validation, allowing the system to incorporate advances in the field without requiring a full model rebuild.

Values-Integrated Factor Research

A significant part of our factor research focuses on signals derived from how companies are run: how they govern capital allocation, manage operational risk, and position themselves relative to regulatory and resource constraints. This is not a screening exercise. We are not removing companies from a universe based on category membership. We are building quantitative signals from underlying data about company behavior, and those signals happen to be closely tied to what most investors would recognize as long-term business quality.

For clients with explicit mission-aligned or values-based mandates, we can design the model to incorporate hard constraints alongside the factor signals. For clients who simply want access to the research without a values overlay, the factors are available on their own merits as components of a broader quantitative program.

Governance as a Quality Signal

How a company governs itself is one of the more analytically tractable sources of forward-looking signal available in public data. Board composition and independence, executive compensation structure relative to long-term performance, audit quality, capital allocation discipline, and proxy voting records all generate measurable, forecastable signals. Poor governance has historically been associated with earnings quality deterioration, regulatory exposure, and capital destruction. We build proprietary governance factors from underlying data rather than from commercial composite scores, which are themselves commoditized and backward-looking.

Operational and Resource Risk

Carbon exposure, physical climate risk, transition risk, and resource efficiency are increasingly relevant to institutional investors managing long-duration portfolios. These are not values-based inputs in our model. They are risk factors: measurable exposures with regulatory, reputational, and financial consequences that show up in returns over the time horizons we care about. We construct these signals from reported data, estimated exposures, and third-party verification signals, and treat them the same way we treat any other factor: evaluate the signal quality, validate against appropriate benchmarks, and include or exclude based on the evidence.

Workforce and Supply Chain Signals

Labor practices, supply chain management, product safety, and customer satisfaction generate data signals in specific sectors and time horizons. We approach this area with appropriate discipline about data quality. The signal-to-noise ratio varies considerably across sectors, and we focus on areas where the evidence for predictive value is demonstrably positive rather than applying broad social data inputs indiscriminately.

How Integration Works

These factors can enter a quantitative model in two ways: as alpha signals that contribute to return forecasts alongside fundamental and technical factors, or as hard constraints that define the investment universe before the model runs. Treating them as alpha signals is the more sophisticated approach and the one we recommend for most clients. Hard constraints are appropriate when the client has explicit mandate requirements that need to be enforced at the portfolio level. Stance has experience building both and can design the integration to match the client's specific requirements.

How a Consulting Engagement Works

A Stance consulting engagement follows a structured process from objective-setting through ongoing model support. The relationship is designed as a multi-year partnership, not a one-time project, because quantitative models require continuous maintenance, data updates, and refinement to remain effective.

Defining the Objective

Every engagement begins with a clear statement of what kind of edge the client is trying to build: differentiating an existing systematic strategy, building a new proprietary model from scratch, integrating sustainability signals into an existing quantitative framework, or adding alternative data to a model that currently relies on standard inputs. The answers shape the entire research design.

Phase 1: Factor Discovery and Data Sourcing

We map the factor space: identifying which signals are available off-the-shelf and should therefore be augmented or replaced, which can be derived from standard data using proprietary methods, and which require alternative data procurement. For clients with existing data vendor relationships, we assess what can be leveraged. For new data needs, we manage the sourcing process and create model-ready features from raw alternative data through the feature engineering pipeline.

Phase 2: Model Construction and Ensemble Building

With the factor library in place, we construct the ML ensemble: selecting learning algorithms, defining variable selection approaches, setting time-horizon parameters, and calibrating the blending methodology. The full pipeline, from data ingestion through feature creation through model training through final forecast generation, is automated end to end.

Phase 3: Performance Assessment and Validation

Model validation follows a structured four-step assessment process. First, standard Fama-French factor alpha benchmarking in a long-only context. Second, Stance's internal framework covering security selection performance and factor signal quality versus random and statistical benchmarks. Third, backtesting on portfolio selection, final portfolio performance, and attribution versus historical benchmarks. Fourth, factor contribution attribution using statistical tooling to evaluate individual factor performance within the model.

Backtested and historical signal results are used to assess model construction quality, not as representations of future performance. All validation is subject to the limitations of historical data, including survivorship bias and regime change. Results are presented as evidence of signal quality, not as performance projections.

This process is designed to establish whether the model has demonstrated alpha-generating characteristics before it manages actual assets.

Phase 4: Ongoing Monitoring, Refinement, and Partnership

A quantitative model is a living system. Our ongoing partnership includes incremental factor sourcing and creation as new data becomes available, model updates and adjustments in response to regime changes or signal decay, technological monitoring to ensure the system is performing as intended, and periodic collaborative sessions with the client's investment team. The relationship is designed to keep the model current, not just operational.

IP and Licensing

The research, factors, and models developed during an engagement are licensed to the client for use in their investment program. License structures are multi-year agreements with ongoing support included, ensuring the client has access to model updates and research as the program evolves. Terms are flexible based on client requirements, including exclusivity arrangements for factors or models developed specifically for a client's mandate.

Risks and Honest Limitations

Overfitting and model fragility. The most common failure mode in quantitative model building is overfitting: constructing a model that performs well in-sample but fails on new data. Our validation process is specifically designed to detect and address overfitting, but no process eliminates this risk entirely.

Data quality and survivorship bias. Alternative data is often noisier than traditional financial data. Historical datasets can carry survivorship bias, changes in vendor methodology, or collection anomalies that produce signals in backtests that do not persist live. We apply rigorous data cleaning and bias correction, but clients should expect that some signals identified in research will not survive deployment.

Factor crowding as strategies scale. Any strategy that works attracts capital. As more institutions identify and trade on the same signals, factor returns compress. Alternative data factors take longer to become crowded because access barriers are higher, but they are not immune. Ongoing factor development is required to stay ahead of the market.

The limits of backtesting. Backtesting is an essential validation tool and an imperfect one. Historical factor performance does not guarantee future performance. We present backtests as evidence of signal quality, not as performance projections.

Ongoing maintenance is required. A quantitative model cannot be built once and left to run. Data vendors change coverage and methodology. Market regimes shift. Factors decay. Our engagement model is structured as a multi-year partnership specifically because the maintenance requirement is real and continuous.

Starting the Conversation

The first conversation is a 45-minute session covering your current investment program, the kind of differentiation you are seeking, and whether our capabilities are a fit. We are direct about cases where our approach is not the right answer.

If you are an advisor, institutional investor, foundation, or family office thinking seriously about building proprietary quantitative research capabilities, whether values-integrated, fundamental, or alternative data-driven, we would like to talk.

Schedule a conversation.

Frequently Asked Questions

Who owns the IP developed during an engagement?

Factors and models developed during a consulting engagement are licensed to the client for use in their investment program. License structures are multi-year agreements with ongoing support. Exclusivity arrangements are available for factors or models developed specifically for a client's mandate.

How long does it take to build a working quantitative model?

A foundational factor model, from scoping through factor library construction, ML model build, and validation, typically takes six to twelve months. The timeline depends on the complexity of the factor space, the number of alternative data sources being integrated, and the client's existing infrastructure.

Can Stance integrate with our existing data infrastructure?

Yes. We assess what can be leveraged from existing vendor relationships before recommending new data sourcing. We have experience working with a range of data environments and custodial infrastructure.

Can we incorporate values-based criteria into a quantitative model without sacrificing returns?

Evidence suggests that well-constructed factor integration around governance and operational risk need not require sacrificing returns, though results vary by strategy, time period, and market environment. The key is treating these as model inputs rather than pure exclusion screens. Past results are not indicative of future performance.

What does ongoing support include?

Model monitoring, data pipeline maintenance, factor updates as new data becomes available, model recalibration when warranted by market conditions or signal decay, statistical performance reporting, and periodic collaborative sessions with the client's investment team. Specific scope is defined in the license agreement.

What happens if a factor stops working?

Factor decay is a normal feature of quantitative investing. Our monitoring framework is designed to detect when a factor's predictive power is degrading. We evaluate whether the issue is cyclical or structural, and in cases of structural decay, we initiate development of a replacement factor as part of the ongoing engagement.