The Scientific Research Approach to Stock Indicators

Most retail investors treat stock indicators as signals. Moving average crossover? Buy signal. RSI below 30? Oversold. P/E under 15? Cheap.

This is backwards. The right frame is to treat every indicator as a hypothesis — a claim about the world that must survive empirical scrutiny before you bet real money on it.

What Makes a Good Scientific Hypothesis?

A scientific hypothesis must be:

Falsifiable — there must be conditions under which it could be proven wrong
Specific — not “this usually works” but “this works under conditions X, Y, Z”
Grounded in mechanism — a causal story for why it should work, not just historical correlation

Stock indicators fail all three tests when used naively. The chart-reader who says “this double-bottom pattern is bullish” usually cannot tell you: under what conditions it fails, what market regime it requires, or why buyers specifically appear at that price level.

The Replication Crisis Applies to Finance

Psychology and nutrition science had their replication crises. Finance has its own — it just moves slower because the feedback loop is a decade rather than a semester.

Campbell Harvey and colleagues (2016) reviewed 316 published factors that “predict” stock returns. After adjusting for multiple testing, most lose statistical significance. Every additional researcher trying new combinations on the same historical data inflates the false discovery rate.

The implication: almost every indicator that looks good in backtests is partly an artifact of data mining. The rigorous researcher must:

Hold out data that was never touched during hypothesis generation
Adjust significance thresholds for the number of hypotheses tested
Require an economic mechanism — not just correlation — before trusting a signal

How to Evaluate an Indicator Scientifically

Step 1: Articulate the Mechanism

Before looking at returns, ask: why should this indicator work?

Price/Earnings ratio: The story is clear. Stocks priced at lower multiples of earnings require less growth to justify the price, leaving a wider margin of safety when growth disappoints. This is mechanistically sound.

RSI at 30: The story is murkier. Mean reversion? Why 30 and not 28 or 35? If it’s measuring oversold momentum, what’s the market microstructure reason a specific threshold creates predictable reversals? Without a mechanism, you’re curve-fitting.

Step 2: Define the Population

An indicator might work brilliantly for large-cap US equities and fail for small-caps, or work in bull markets and destroy capital in bear markets. Specify:

Market cap range
Sector applicability
Market regime (trend vs. range-bound)
Time horizon

Vague: “The golden cross works.”
Scientific: “A 50/200 SMA crossover produces positive excess returns in large-cap US equities over 12-month holding periods in bull markets, but not during secular bear markets.”

Step 3: Stress-Test With Out-of-Sample Data

Split your historical data. Develop the hypothesis on the first half. Test — only once — on the second half. If you adjust the parameters after seeing the second-half results, you’ve contaminated your test.

Better still: use a dataset you genuinely haven’t seen. International markets, a different time period, or asset class. Robust signals generalize; overfit ones collapse.

Step 4: Account for Transaction Costs and Taxes

Academic papers report gross returns. Real portfolios pay spreads, commissions, short-term capital gains taxes, and suffer market impact at scale. A signal generating 2% annualized gross alpha often produces 0% net alpha after costs.

Run the numbers honestly.

Indicators That Survive Scrutiny

The following have reasonably strong evidence and defensible mechanisms:

Value (P/E, P/B, P/FCF)
Half a century of data across multiple international markets. The mechanism — investor over-extrapolation of recent growth — is behaviorally grounded. Weakened but not eliminated in the era of intangible-heavy business models.

Earnings Momentum
Stocks with recent positive earnings surprises tend to continue outperforming over the next 3–6 months (post-earnings announcement drift). Mechanism: analysts and investors underreact to new information, and price discovery is gradual.

Quality Factors (low debt, high ROIC, stable earnings)
Buffett’s track record is a 60-year single data point, but it aligns with quantitative evidence. High-quality companies compound better and crash less. The mechanism is fundamental: durable competitive advantages reduce business risk.

Low Volatility Anomaly
Counterintuitively, low-beta stocks outperform high-beta stocks on a risk-adjusted basis over long periods. Mechanism: institutional constraints (benchmarking, leverage limits) push capital away from low-volatility assets, creating mispricing.

Red Flags in Indicator Claims

Watch for these patterns:

Optimized parameters with no economic rationale (why 14-period RSI and not 13?)
No mention of transaction costs
Cherry-picked time periods (2010–2020 bull market only)
No out-of-sample test — only in-sample backtest
No market regime analysis — does it work in drawdowns?
“It just works” — without a causal story

The Analyst Mindset

The best fundamental analysts think like scientists in court — they’re building a prosecution brief, not a research paper. They ask: what is the strongest possible case for the other side? What would have to be true for this thesis to be wrong?

That’s the posture that makes indicators useful. Not as oracles, but as structured ways to surface questions you then answer with primary research.

A stock screener tells you where to look. The scientific method tells you whether what you’re looking at is real.

At Reigraph, every signal in the app is accompanied by the underlying data and context — so you can evaluate the reasoning, not just accept the output. That’s the only honest way to do this.