Same Signal, Opposite Result: How Stock Type Determines Whether Technical Indicators Work
We applied the three strongest technical indicators from our ETF study to NVDA, NFLX, WMT, COST, BAC, and JNJ — and found that the direction of the signal completely reverses depending on whether the stock trends or mean-reverts.
The Follow-Up Question That Changes Everything
In our previous study, we ran 856 statistical tests across 14 technical indicators and 5 ETFs over 20 years (2003-2023). The verdict was clear: mean-reversion signals (RSI < 30, Keltner Channel lower touch, OBV divergence) are statistically real, with permutation p-values below 0.0001. Trend-following signals (SMA crossovers, PSAR, ROC) are noise.
The natural next question: does this hold for individual stocks?
We took the three strongest indicators from that study and ran the same rigorous testing — Welch’s t-test, 5,000-permutation bootstrap, Bonferroni correction — against six individual names: NFLX, NVDA, WMT, COST, BAC, and JNJ. We used data from 2003 to 2023, performing ninety-six hypothesis tests in total.
The result wasn’t what we expected. The signals work — but on some stocks they manifest in reverse.
The Setup
Indicators Tested
The three champions from the ETF benchmark:
- RSI < 30 (Oversold) — 6 Bonferroni survivors in ETF study, best Cohen’s d = 0.302 on SPY
- Keltner Channel lower band touch — 7 survivors, best d = 0.311
- OBV vs. its 20-day SMA — 8 survivors, d = 0.219 on SPY (most consistent)
Stocks Tested
| Ticker | Sector | Character |
|---|---|---|
| NVDA | Semiconductors | High-growth, multi-year uptrend |
| NFLX | Streaming/Tech | High-growth, high-volatility |
| WMT | Consumer Staples | Defensive, steady cash flows |
| COST | Consumer Staples | Defensive, membership model |
| BAC | Financials | Rate-sensitive, macro-driven |
| JNJ | Healthcare | Defensive, dividend grower |
Methodology
We calculated log forward returns at 1, 5, 10, and 20 trading days. Permutation p-values were generated using 5,000 shuffles. The Bonferroni correction was applied across all 96 tests, with a threshold of α = 0.0005 for statistical significance. Cohen’s d effect sizes were determined for each test to quantify the magnitude of the observed differences.
Results: The Direction Splits by Stock Type
Total Bonferroni Survivors: 18 out of 96
Expected by chance: 4.8. We obtained 18 — indicating clearly detectable signal. However, when observing which tests survived, a sharp pattern emerges.
| Stock | Survivors | Best Cohen’s d | Signal Direction |
|---|---|---|---|
| NVDA | 5 | −0.459 | Negative — signals predict continuation |
| NFLX | 4 | −0.416 | Negative — signals predict continuation |
| JNJ | 3 | +0.308 | Positive — mean-reversion works |
| WMT | 3 | +0.270 | Positive — mean-reversion works |
| COST | 2 | +0.374 | Positive — mean-reversion works |
| BAC | 1 | −0.200 | Negative — rate dynamics override signals |
Specific permutation p-values for significant results ranged from 0.0000 to 0.0003, supporting the statistical robustness of the findings.
Defensive Stocks: Mean-Reversion Works as Expected
On WMT, COST, and JNJ, the signals behave exactly as they did on SPY and GLD in the ETF study.
Top Performing Tests (Positive Direction)
| Stock | Signal | Horizon | Edge (bps) | Cohen’s d | p-value |
|---|---|---|---|---|---|
| COST | RSI < 30 | 10d | +152 | +0.374 | 0.0003 |
| COST | RSI < 30 | 20d | +185 | +0.328 | 0.0002 |
| JNJ | KC lower touch | 10d | +95 | +0.308 | 0.0004 |
| JNJ | KC lower touch | 20d | +121 | +0.287 | 0.0001 |
| JNJ | KC lower touch | 5d | +56 | +0.255 | 0.0005 |
| WMT | RSI < 30 | 20d | +133 | +0.270 | 0.0003 |
| WMT | RSI < 30 | 10d | +78 | +0.221 | 0.0004 |
COST is the standout defensive name. RSI < 30 fires 89 times in 20 years and produces +185 bps over the following 20 days with d = 0.374. That’s a larger effect size than RSI on SPY (d = 0.302). Costco’s membership model and resilient fundamentals mean true oversold conditions are temporary and reliable buying opportunities.
JNJ is the strongest KC result among individual stocks. The healthcare giant’s low volatility and defensive cash flows create a mean-reverting price dynamic: when JNJ touches the lower Keltner band, it rebounds. Three separate (ticker × horizon) combinations survive Bonferroni correction, with Cohen’s d values between 0.25 and 0.31.
WMT shows RSI < 30 working cleanly across horizons, though only the 20-day version survives Bonferroni correction. The 10-day result (d = 0.22) is raw-significant and practically meaningful.
Growth Stocks: The Same Signals Predict Further Decline
On NVDA and NFLX, every Bonferroni-surviving test shows a negative edge — meaning the oversold signal predicts the stock will continue falling, not recover.
Top Performing Tests (Negative Direction)
| Stock | Signal | Horizon | Edge (bps) | Cohen’s d | p-value |
|---|---|---|---|---|---|
| NVDA | KC lower touch | 20d | −633 | −0.459 | 0.0000 |
| NVDA | RSI < 30 | 20d | −581 | −0.419 | 0.0001 |
| NFLX | RSI < 30 | 20d | −631 | −0.416 | 0.0001 |
| NFLX | KC lower touch | 20d | −398 | −0.262 | 0.0004 |
| NFLX | RSI < 30 | 10d | −338 | −0.317 | 0.0002 |
| NVDA | KC lower touch | 10d | −212 | −0.218 | 0.0005 |
These are not marginal effects. A Cohen’s d of −0.46 is a large and robust result. When NVDA touches the Keltner lower band, the average 20-day forward return is 633 basis points below what non-signal days produce, with permutation p-value of 0.0000.
The mechanism is the inverse of mean-reversion: momentum and trend continuation. NVDA has compounded at extraordinary rates since 2016, with violent multi-month corrections that regularly pushed RSI and KC into oversold territory — before continuing lower. The 2022 drawdown saw NVDA fall 66%. RSI hit oversold levels dozens of times on the way down. Anyone buying those “oversold” signals kept buying a falling knife.
NFLX shows the same pattern. The stock’s crash from $700 to $160 in 2022 generated oversold signals for months while the price halved again. Its structural volatility means “oversold by RSI” is a much weaker signal about fair value than it is for a steady compounder like Costco.
OBV on Individual Stocks: Signal Weakens but Partially Holds
OBV was the most consistent ETF signal with 8 Bonferroni survivors and d = 0.219 on SPY. On individual stocks it largely disappears — but two names produce surviving tests.
NVDA OBV at 20-day horizon: Both directions survive (above SMA = +164 bps, below SMA = −164 bps), d = 0.119, p-value = 0.0005. This is OBV working in its correct direction on NVDA — consistent with the trend: when volume is bullish, NVDA outperforms; when volume is bearish, it underperforms. Not a mean-reversion play, but an OBV-as-trend-confirmation play.
WMT OBV below SMA: Two survivors (10d and 20d), d = 0.094 and 0.140, with p-values of 0.0004 and 0.0002 respectively. When WMT’s OBV is below its moving average, subsequent returns are positive (+34 bps at 10d, +69 bps at 20d). Same mean-reversion dynamic as the ETF study.
Everywhere else: OBV Cohen’s d values hover near zero for NFLX, COST, BAC, and JNJ. The volume signal that works on a broad market ETF does not reliably translate to individual stock dynamics.
BAC: When Macro Overrides Everything
Bank of America shows only one Bonferroni survivor: KC lower touch at 10 days, −170 bps, d = −0.200, with p-value = 0.0005. This is a negative result — a KC lower touch on BAC predicts underperformance over the following two weeks.
Banks are driven by interest rate dynamics, credit cycles, and systemic risk events (2008, 2023 regional banking crisis) that no technical indicator is built to capture. RSI < 30 on BAC happens at exactly the moments when the macro picture is most uncertain — and the stock may well continue falling for fundamental reasons. The 10-day KC result appears to capture this: panic selling in financials is more often trend continuation than buying opportunity.
The Unifying Framework
The results suggest a single principle that explains all the findings:
Technical mean-reversion signals work on stocks whose price reflects fundamentals that are stable and predictable. On stocks where the price reflects growth expectations, “oversold” means the narrative has broken — not that the price has strayed from fair value.
COST and JNJ have earnings streams that grow steadily. When macro fears or broad market selloffs push them to Keltner lower bands, they are genuinely cheap relative to their intrinsic value, and they recover.
NVDA and NFLX are priced on expectations. When they sell off hard enough to trigger RSI < 30, it often means the market is revising those expectations — a process that can take months and many further percentage points of decline before it stabilizes.
This isn’t a novel insight in fundamental analysis. But it’s rarely stated as a quantitative rule for which technical signals to use on which stocks. Now it is.
Practical Decision Framework
Use this as a pre-filter before applying any mean-reversion indicator:
Mean-reversion signals are applicable when:
- The stock has consistent earnings and cash flows (consumer staples, healthcare, utilities)
- P/E ratio is reasonable relative to sector (not a high-multiple growth name)
- The stock has not been in a persistent multi-year trend
- The oversold condition appears to be macro-driven, not company-specific
Mean-reversion signals should be avoided or inverted when:
- The stock is a high-growth, high-multiple name (tech, semis in uptrend)
- The selloff is narrative-driven (earnings miss, guidance cut, competitive threat)
- The stock was in a parabolic run before the selloff
- RSI < 30 is occurring for the 3rd or 4th time in 6 months
For OBV: Use it as trend confirmation on momentum stocks (NVDA-type), not as a mean-reversion signal. Use it as a mean-reversion pre-filter on steady compounders (WMT-type) only.
Entry and Exit Rules
- Entry: Upon indicator trigger (e.g., RSI < 30), initiate a position at the next day’s open.
- Exit: Exit after the designated holding period (1, 5, 10, or 20 days), or if the indicator reverses (e.g., RSI rises above 50).
Impact of Transaction Costs
All results are pre-transaction-cost. Bid-ask spreads on individual stocks, particularly in size, might erode edges. The impact of slippage should be factored into potential real-world applications.
Limitations
20-year average includes multiple regimes: NVDA’s extraordinary run from 2016 to 2024 heavily weights the results. In an environment where NVDA underperforms for years, these signals might behave differently. We cannot know which regime we’re currently in.
Individual stock results have fewer signal events: COST RSI < 30 fires only 89 times across 20 years. Statistical power is lower than ETF results; effect size estimates have wider confidence intervals.
These signals are tested in isolation: A COST RSI < 30 signal that also coincides with broad market stress (SPY RSI < 35) may have different properties than one occurring during a stable market. Multi-condition filters are a natural follow-on.
Transaction costs: Pre-cost edges. Bid-ask spreads on individual stocks, particularly in size, will erode edge.
Summary
| Stock | RSI < 30 Works? | KC Lower Touch Works? | OBV Works? | Direction |
|---|---|---|---|---|
| COST | ✓ (d=0.37) | ✓ (d=0.28) | — | Positive |
| JNJ | ✓ (d=0.28) | ✓ (d=0.31) | — | Positive |
| WMT | ✓ (d=0.27) | — | ✓ (d=0.14) | Positive |
| NVDA | ✗ flips (d=−0.42) | ✗ flips (d=−0.46) | ✓ trend-only | Negative |
| NFLX | ✗ flips (d=−0.42) | ✗ flips (d=−0.26) | — | Negative |
| BAC | — | ✗ flips (d=−0.20) | — | Negative |
The ETF study told us which indicators work. This study tells us where they work. The answer is not uniformly across all stocks — it depends entirely on whether the stock’s price is driven by fundamentals that revert or by narratives that trend.
This is Part 2 of the Reigraph Research Technical Indicator Series. Read Part 1: Do Technical Indicators Actually Work? We Ran 856 Statistical Tests to Find Out
Reigraph Research · May 2026
This study is not investment advice. Statistical edge does not guarantee trading profit. All results are pre-transaction-cost and based on historical data.