Why We Stopped Building a Strategy and Started Building an Engine
How a small-cap quant system evolved from a single backtest into a configurable factor platform — and why that distinction matters for Indian portfolio managers.
Most quant projects in India follow a familiar arc. You find an edge — maybe momentum works in small caps, or promoter buying predicts returns. You build a model around that edge. You backtest it. The numbers look good. You deploy capital.
Then the edge decays. Or the market regime shifts. Or you realize the strategy works beautifully in bull markets and bleeds in sideways ones. So you build another model. And another. Each one is a standalone system with its own data pipeline, its own scoring logic, its own portfolio construction rules. Before long, you're maintaining three disconnected systems that share 80% of their infrastructure but can't talk to each other.
We went through this exact cycle with RevDog. And the breakthrough wasn't a better factor or a cleverer signal. It was recognizing that we were building the wrong thing.
The Strategy Trap
RevDog started as SmallCap Catalyst — a regime-aware factor scoring system for the NIFTY SmallCap 250. It scores stocks across fundamental factors (quality, governance, valuation, growth), gates which factors are active based on macro conditions, and constructs portfolios with dynamic cash allocation. In backtesting from November 2022 to November 2025, it delivered 18.1% CAGR with a maximum drawdown of -15.3%, compared to the benchmark's -26.1% drawdown. The trade-off was intentional: give up 2.6 percentage points of annual return to cut drawdowns by 11 points.
Then we built SmallCap Dislocation — a mean-reversion overlay that detects forced selling in small caps and trades the recovery. Completely different philosophy. Event-driven, not factor-driven. Daily scanning, not quarterly rebalancing. Fixed position sizes, not Kelly-based tiers. It uses Catalyst's idle cash, improving capital efficiency from 67% to 91.5% deployed.
Here's what we noticed: roughly 70% of the infrastructure was identical across both strategies. Regime detection — the same. Governance screening — the same. Data pipelines, eligibility filtering, decision logging — all the same. What differed was which factors to score, how to size positions, and when to rebalance.
We had accidentally built most of a general-purpose factor engine. We just hadn't designed it as one.
What a Factor Engine Actually Means
Dimensional Fund Advisors manages over $700 billion using what is essentially a single engine configured differently across dozens of funds. Their US Small Cap Value fund and their International Core Equity fund share the same scoring infrastructure, the same portfolio construction framework, and the same execution system. What differs is the universe (US small caps vs. international), the factor weights (heavier value tilt vs. balanced), and the construction parameters (concentration limits, sector caps, turnover budgets).
The insight is simple but powerful: the engine is the product. The strategies are configurations.
When Dimensional improves their execution logic — say, better algorithms for patient trading in illiquid names — every fund benefits simultaneously. When they refine their factor scoring methodology, all strategies get smarter at once. The improvement compounds across the entire platform, not just one fund.
We asked ourselves: what would it take to make RevDog work the same way?
RevDog's Engine Architecture
The engine has six layers. Each layer does one job and doesn't know which strategy is calling it.
Regime Detection reads five macro signals — India VIX, USD/INR movement, small-cap relative performance, gold versus equity flows, and Bank NIFTY relative strength. It outputs a regime state ranging from Crisis to Bullish, with a cost-relief modifier when falling crude prices signal margin expansion for Indian manufacturers. This layer is fully shared. Every strategy on the platform sees the same macro environment.
Eligibility Filtering applies configurable rules to narrow the universe. For Catalyst, this means profitability gates (positive ROE, positive PAT), leverage limits (debt-to-equity below 8), liquidity thresholds (minimum daily turnover of ₹1 crore), and IPO seasoning (at least 12 months since listing). A different strategy might use the same engine with different thresholds — a more aggressive approach could lower the liquidity bar, while a large-cap strategy would tighten it.
Governance Screening applies India-specific risk filters that don't exist in any global factor framework. Promoter stake declining by 25 percentage points in 12 months? Score capped at 3.0 — effectively an exit signal. Promoter pledge exceeding 80%? Same treatment. Two consecutive years of negative operating cash flow? Score capped at 4.5, forcing the position into a reduced allocation. These rules are calibrated from real Indian market blowups — the Yes Banks, the DHFLs, the Manpasand Beverages incidents that turn a 5% portfolio position into a -90% loss.
Factor Scoring is where strategies diverge most — but the scoring engine itself is shared. It takes a list of factors, a set of weights, and a normalization method as inputs. Catalyst's configuration specifies six fundamental factors (quality, governance, valuation, growth, capital efficiency, income safety) with sector-relative normalization — meaning a bank with 15% ROE is scored against other banks, not against pharma companies with naturally higher returns on equity. The weights shift by regime: in stress, only governance and income safety factors remain active. In normal markets, quality and growth factors come online. In risk-on environments, the full factor set activates.
Portfolio Construction is the one layer that varies structurally between strategies. Catalyst uses Kelly-based position sizing in conviction tiers — high-conviction positions get 6-8% allocation, medium 4-6%, low 2-4%, all adjusted by regime. Positions decay gradually as conviction weakens, with a hysteresis rule preventing whipsaw exits. Dislocation uses fixed 3.5% positions with hard profit targets and time stops. Both constructors apply a liquidity constraint: no position can exceed 20% of the stock's 90-day average daily turnover, ensuring the backtest reflects achievable real-world execution.
Monitoring generates structured decision traces for every portfolio action — not just what the system did, but why. Stock excluded? The trace records whether it was score-based, diversification-limited, or governance-capped. Position resized? The trace records whether it was a decay downgrade, sector cap hit, or regime-driven exposure change. This audit trail exists for every rebalance, every strategy, and it's queryable: "Why was Stock X excluded on Date Y?" has a precise, reproducible answer.
Why This Matters for Portfolio Managers
If you run a PMS or manage HNI portfolios, you face a specific set of problems that a strategy can't solve but an engine can.
Different mandates need different configurations, not different systems. A wealth preservation mandate for a pre-retirement HNI needs conservative regime thresholds, lower position concentration, and capital preservation emphasis. An aggressive growth mandate for a younger investor needs higher equity allocation in risk-on environments, tolerance for larger drawdowns, and growth factor emphasis. With an engine, these are parameter changes — not separate systems to build and maintain.
Factor research compounds. When we discovered that ROE and ROCE flip to negative predictive power during market stress — a finding from walk-forward validation across 15-month rolling windows — that insight immediately benefited every strategy running on the engine. We didn't need to re-discover it separately for each mandate.
Governance rules protect all portfolios equally. The promoter pledge screening that would have flagged a Yes Bank or a DHFL runs identically across every strategy configuration. Building this once, rigorously, is better than building it three times with slight inconsistencies.
Auditability scales. SEBI's increasing focus on transparency and explainability in portfolio management isn't going away. Having structured decision traces for every portfolio action across every strategy — generated automatically, not retrofitted for compliance — is infrastructure that becomes more valuable over time. When a client asks why a stock was sold, the answer isn't "the PM decided" — it's a traceable chain from macro regime detection through factor scoring to portfolio construction constraints.
New strategies are weeks, not months. Once the engine exists, launching a midcap strategy means writing a configuration file that specifies: universe (NIFTY MidCap 150), eligibility rules (perhaps higher liquidity thresholds), factor weights (perhaps heavier momentum tilt), and construction parameters (perhaps more positions with lower concentration). The regime detection, governance screening, scoring engine, and monitoring infrastructure are inherited. The marginal cost of a new strategy approaches the cost of the research to define it — the engineering is already done.
What We're Not
We're not trying to replace discretionary portfolio management. The engine doesn't pick themes or time sectors or make macro calls beyond its regime detection framework. It doesn't replace a PM's judgment about whether the defense sector rally has legs or whether a specific management team is trustworthy.
What it replaces is the undifferentiated heavy lifting: screening 250 stocks quarterly, normalizing factor scores across sectors, enforcing position limits and governance checks, generating audit trails, adapting exposure to market conditions. The work that's essential, repetitive, error-prone when done manually, and identical across mandates.
The PM's job becomes higher-value: deciding which configuration matches which client's mandate, overriding the system when qualitative information (a management meeting, an industry insight) warrants it, and evaluating whether the engine's factor assumptions still hold.
Where We Are Today
SmallCap Catalyst is live with real capital. The backtest covers November 2022 through November 2025, validated using walk-forward methodology (not just a single in-sample optimization). Performance after transaction costs: 18.1% CAGR, 0.84 Sharpe, -15.3% maximum drawdown, with 49% better Calmar ratio than the benchmark.
SmallCap Dislocation runs as an overlay, using Catalyst's idle cash. The combined system improves CAGR by 4.3 percentage points while maintaining the same maximum drawdown — primarily by deploying capital that would otherwise sit as cash.
The engine-strategy separation is underway. Factor scoring is being refactored with sector-relative normalization. Eligibility filtering is moving to configurable rules. The monitoring layer is being extended with factor drift detection. By the time these changes ship, adding a new strategy configuration becomes a research question, not an engineering project.
The Opportunity in Indian Markets
Here's the uncomfortable truth about systematic investing in India: the infrastructure doesn't exist.
In the US, a retail investor can access Dimensional's factor engine through ETFs at 0.25% expense ratio. They get sector-relative scoring, liquidity-aware sizing, patient execution, and decades of validation — for the price of a passive fund. India's best factor ETFs use the NIFTY index methodology: binary inclusion/exclusion, simple ranking, fixed semi-annual rebalancing, no sector normalization, no governance screening. They're version 1.0 products in a market that's ready for version 3.0.
For PMS firms, this gap is both a challenge and an opportunity. The challenge: building this infrastructure in-house takes 2-3 years and significant engineering investment, with no guarantee the team has the quant + engineering + India-market expertise to get it right. The opportunity: whoever deploys institutional-grade systematic infrastructure in Indian small and mid caps — with the governance layer, the regime awareness, and the auditability that Indian markets specifically demand — captures a market that's currently underserved.
That's what we're building. Not a strategy. An engine.