How to Screen Stocks: Building a Systematic Investment Process

Why Most Stock Screening Processes Fail

Most investors screen stocks badly. They open Finviz, set a P/E under 15 and market cap over $10 billion, scroll through the results, and call it analysis. This is not a process. It is a recipe for buying value traps, missing compounders, and generating the same mediocre returns as the index — before fees.

The problem is not the tools. Bloomberg Terminal, FactSet, Capital IQ, and even free platforms like Finviz and Stock Rover offer powerful screening capabilities. The problem is that investors confuse screening with stock picking. Screening is the first stage of a multi-step process. Its job is to shrink the investable universe from thousands of names to a manageable list of candidates that share characteristics historically associated with outperformance. That is all. The screen does not tell you what to buy. It tells you where to look.

We have seen this failure pattern repeatedly across buy-side firms, RIAs, and individual investors. An analyst builds a screen that returns 20 names, picks the three she recognizes, and ignores the rest. Or a PM adjusts screening criteria until his current holdings show up in the results, validating existing positions instead of finding new ones. Or — the most common failure — someone backtests 50 different factor combinations until one shows 25% annualized returns over the last decade, then discovers it was entirely overfitted when forward returns revert to the mean.

A properly constructed systematic screening process avoids all of these traps. Here is how to build one.

Step 1: Define Your Universe and Factor Framework

Before selecting a single screening criterion, you need two things: a clearly defined investable universe and a factor framework grounded in economic logic, not data mining.

Defining the Universe

Your universe should reflect your mandate constraints. A US large-cap equity manager screens the S&P 500 or Russell 1000. A global all-cap manager might screen the MSCI ACWI Investable Market Index, covering roughly 9,000 names across 23 developed and 24 emerging markets. A small-cap specialist screens the Russell 2000 or S&P 600.

For individual investors without mandate constraints, we recommend starting with the Russell 3000, which covers approximately 98% of the US equity market. This is broad enough to capture opportunities across the capitalization spectrum but focused enough to avoid the data quality issues that plague micro-cap and international databases. Apply a minimum liquidity filter — average daily dollar volume above $2 million — to ensure you can actually execute trades without excessive market impact.

Building the Factor Framework

A factor is any measurable company characteristic that has a theoretical and empirical basis for predicting future returns. The academic literature, starting with Fama and French's three-factor model in 1993 and expanding through Carhart (momentum), Novy-Marx (profitability), and Asness (quality minus junk), has identified dozens of factors. Most are redundant or spurious. Focus on the five with the strongest economic intuition and out-of-sample evidence:

• Quality: Measured by ROIC, ROE, gross profitability (gross profit / total assets), or operating margins. High-quality businesses generate more economic value per dollar of invested capital and compound intrinsic value faster. Joel Greenblatt's “Magic Formula” uses ROIC as its primary quality metric. Novy-Marx demonstrated that gross profitability alone has as much power as traditional value factors.
• Value: Measured by EV/EBIT, P/FCF, earnings yield (EBIT/EV), or price-to-book. We prefer enterprise value-based metrics over price-based metrics because they account for differences in capital structure. A company with a low P/E but massive debt may not be cheap on an EV/EBIT basis. EV/EBIT is our preferred primary value factor.
• Momentum: Measured by 6-month or 12-month total return minus the most recent month (to strip out short-term reversal effects). Momentum captures behavioral persistence — investors underreact to positive information and slowly update their priors. Jegadeesh and Titman's 1993 study remains one of the most replicated findings in financial economics.
• Financial Health: Measured by Altman Z-Score, interest coverage ratio, net debt/EBITDA, or Piotroski F-Score. This screens out companies that may look cheap or profitable on a trailing basis but face solvency risk. Piotroski's F-Score, a nine-point checklist of profitability, leverage, and operating efficiency, has been shown to improve returns within value stock portfolios by 7.5% annually.
• Earnings Quality: Measured by accruals ratio (change in net operating assets / total assets), cash conversion (operating cash flow / net income), and revenue recognition patterns. High-accrual companies — those where reported earnings significantly exceed cash flows — underperform by 5–10% annually on average, a phenomenon first documented by Sloan in 1996 and still exploitable today.

Step 2: Construct the Composite Score and Set Thresholds

Single-factor screens are seductive but fragile. A pure value screen in 2020–2021 would have loaded you into energy, financials, and airlines — cheap for a reason — while missing Nvidia, ASML, and every quality compounder that drove market returns. A pure momentum screen in January 2022 would have put you into the most speculative growth names right before the Fed hiking cycle destroyed them.

Multi-factor composite scores are far more robust. Here is how to build one:

First, normalize each factor to a percentile rank within your universe. If you are screening 3,000 stocks, the company with the highest ROIC gets a quality score of 100, the lowest gets 0, and everyone else is ranked linearly between. Percentile ranking is superior to z-score normalization because it is insensitive to outliers — a company with 200% ROIC (likely a data error) would distort z-scores but only occupies the 100th percentile in a rank-based system.

Second, weight the factors based on your investment philosophy. We recommend:

Third, set a threshold. We recommend passing the top quartile (top 25%) of composite scores through to the next stage. For a 3,000-stock universe, this yields roughly 750 names — still too many for deep analysis but appropriate as a first filter. A secondary screen using sector-specific factors or qualitative criteria will narrow this further.

Here is the contrarian take that most screening guides will not give you: we believe quality should be weighted more heavily than value for most investors. Value alone has underperformed for the last 15 years in a low-rate, innovation-driven market. Quality plus value — what we call “cheap quality” or what Greenblatt calls the Magic Formula — avoids both value traps and speculative growth. The companies that rank in the top 20% on both ROIC and EV/EBIT have historically outperformed the S&P 500 by 4–6% annually with lower drawdowns.

Step 3: Backtest with Discipline, Not Hope

Backtesting is where 90% of screening processes go wrong. The mechanics are simple: apply your screening criteria to historical data, simulate portfolio construction at regular intervals, and measure returns, drawdowns, and risk-adjusted performance versus a benchmark. The execution is treacherous.

The three deadly sins of backtesting are overfitting, survivorship bias, and look-ahead bias. Overfitting occurs when you test dozens of factor combinations and weight configurations until you find one that shows great historical returns. With enough parameters, you can fit any dataset. The antidote is out-of-sample testing: build your model on data from 2005–2015, then test it on 2015–2025 data that was not used in construction. If performance collapses out of sample, you have overfit.

Survivorship bias occurs when your historical database only contains companies that still exist today. This systematically excludes bankruptcies, acquisitions of failing companies, and delistings — all of which disproportionately affect value and small-cap screens. A backtest without delisted companies overstates value factor returns by 200–400 basis points annually. CRSP, Compustat, and Portfolio123 provide survivorship-bias-free databases. Free tools like Finviz do not.

Look-ahead bias occurs when your backtest uses financial data that was not actually available at the time of the simulated screening. If you screen stocks on January 1, 2020 using full-year 2019 earnings, but those earnings were not reported until February 2020, your backtest has a look-ahead bias. Always use point-in-time data with appropriate reporting lag — typically 60–90 days after fiscal quarter end for SEC-filed financials.

For a detailed guide on incorporating AI into the analysis that follows the screening stage, our article on AI-powered quantitative screening and stock selection walks through the technical implementation.

Step 4: Layer Qualitative Filters on Quantitative Output

The quantitative screen gets you to the 20-yard line. Qualitative analysis scores the touchdown. No screen, no matter how sophisticated, can capture competitive moat durability, management quality, regulatory risk, or industry structure. These require human judgment — or, increasingly, AI-augmented judgment.

After your composite score ranks stocks and the top quartile passes through, apply three qualitative filters:

Competitive Position Assessment. Does the company have a durable competitive advantage? High ROIC in the quantitative screen suggests a moat exists, but it does not tell you whether that moat is widening or narrowing. A company with 25% ROIC and declining market share is very different from one with 25% ROIC and expanding share. Look for network effects (Visa, Mastercard), switching costs (Oracle, SAP), intangible assets (LVMH, Hermes), and cost advantages (Costco, TSMC). We wrote extensively about this framework in our piece on AI-powered competitive analysis for equity research.

Management Capital Allocation. How does management deploy free cash flow? Companies that reinvest at high incremental ROIC, make disciplined acquisitions, and return excess capital through buybacks at attractive valuations create compounding value. Companies that empire-build with overpriced M&A, dilute shareholders with excessive stock-based compensation, or hoard cash at near-zero returns destroy it. Track the five-year history of capital deployment: organic reinvestment, M&A, buybacks, dividends, and debt paydown.

Industry Structure. Is the industry consolidating or fragmenting? Do the top three players control 60%+ of the market (favorable) or less than 20% (challenging)? Is the industry facing secular growth or decline? A high-quality company in a structurally declining industry (print media, traditional tobacco) is a very different investment from a high-quality company in a growing industry (cloud infrastructure, medical devices).

Step 5: Build the Monitoring and Rebalancing Cadence

A screening process is not a one-time event. It is a recurring discipline. The cadence matters more than most investors realize.

We recommend quarterly rescreening, aligned with earnings seasons. Run the screen after Q4 results (late February/early March), Q1 results (late May), Q2 results (late August), and Q3 results (late November). This ensures your screening data reflects the most recent fundamental updates. Between screenings, monitor existing positions for thesis-breaking developments — management changes, competitive disruptions, accounting restatements, or insider selling clusters — using real-time alert systems rather than periodic rescreening.

Rebalancing frequency has a direct impact on returns and costs. Academic research by Novy-Marx and Velikov (2016) shows that quarterly rebalancing captures most of the available factor premium while keeping turnover at manageable levels (40–60% annualized for a quality-value composite). Monthly rebalancing improves factor capture by roughly 50 basis points but increases turnover to 100–150%, which in taxable accounts often destroys the incremental return through short-term capital gains taxes.

For individual investors managing their own accounts, semi-annual rebalancing (twice per year) is likely the optimal frequency. It captures the majority of factor premium, keeps costs low, and imposes a disciplined but not burdensome process. The key is consistency. Skipping a rebalancing because markets feel uncertain is how systematic processes degrade into discretionary ones.

Common Mistakes and How to Avoid Them

After building screening processes for institutional and individual investors over the past decade, we have catalogued the most frequent failure modes:

Using sector-agnostic screens without adjustment. A P/E of 12x is cheap for a software company and expensive for a bank. Screening the entire market on a single absolute threshold produces portfolios heavily concentrated in low-multiple sectors (financials, energy, utilities) while excluding high-ROIC growth sectors entirely. Solution: use sector-relative rankings. A software company at the 20th percentile of sector P/E is the value play, even if its absolute P/E is 25x.

Ignoring the denominator. Revenue growth of 30% means nothing if driven entirely by acquisition. ROIC of 40% is misleading if intangible assets are excluded from invested capital (common in companies with heavy R&D or brand value). Always sanity-check the metrics driving your screen by reading at least one year of financial statements for any stock that passes. For a framework on reading annual reports efficiently, see our guide on how to read an annual report like a professional analyst.

Confirmation bias in qualitative overlay. You screen 100 stocks, then research only the 5 you already know and like. This defeats the purpose of systematic screening. Force yourself to analyze at least 3 names you have never heard of from each screening cycle. Some of the best investments come from companies you did not know existed until the screen surfaced them.

Abandoning the process after underperformance. Every factor goes through drawdown periods. Quality underperformed in 2020–2021 as speculative stocks surged. Value underperformed for a decade from 2010–2020. Momentum crashes happen fast and hard, as in March 2009 and March 2020. If you abandon a well-constructed process after 12 months of underperformance, you will inevitably switch to whatever just worked — buying high and selling low.

Frequently Asked Questions