VWAP + RSI Four-Quadrant Study: Why the Regime Filter Doesn't Work the Way You Think
We tested the theory that stocks above VWAP trend up until overbought, and below VWAP trend down until oversold. 616 hypothesis tests across 11 assets and 20 years reveal a more nuanced — and more actionable — picture.
The Hypothesis
A popular framework in technical trading goes like this:
- If price is above VWAP, the stock has upward momentum — stay long until RSI reaches overbought territory (>70)
- If price is below VWAP, the stock is in a downtrend — stay short until RSI reaches oversold (<30)
It’s a clean, intuitive framework. VWAP acts as the regime filter; RSI signals the exhaustion point. Used together, they’re supposed to tell you both direction and duration.
We tested it the same way we’ve tested every other hypothesis in this series: 20 years of daily data, Welch’s t-test, 5,000-permutation bootstrap, Bonferroni correction across all 616 tests. Eleven assets — five ETFs (SPY, QQQ, GLD, TLT, EEM) plus six individual stocks (NFLX, NVDA, WMT, COST, BAC, JNJ). Three VWAP windows (10, 20, 50-day). Four forward return horizons.
The framework is partially right — but the mechanism is different from what most traders assume, and the best signal it produces is the opposite of what the hypothesis predicts on growth stocks.
The Design: Four Quadrants
Classic intraday VWAP resets each morning and requires tick data. For daily bars we use rolling VWAP — typical price (H+L+C)/3 weighted by volume over a rolling window:
VWAP_N = Σ(typical_price × volume, N days) / Σ(volume, N days)Every trading day falls into one of four quadrants:
| RSI < 70 (room to run) | RSI ≥ 70 (extended) | |
|---|---|---|
| Price > VWAP | Q1: Bull regime, not overbought | Q2: Bull regime, stretched |
| Price < VWAP | Q3: Bear regime, not yet oversold | Q4: Bear regime, oversold |
The user’s hypothesis predicts Q1 forward returns should be strongly positive and Q3 should be strongly negative. We also tested RSI alone (no VWAP) as a control to measure what VWAP adds incrementally.
Finding #1: VWAP Adds Almost No New Information to RSI
The first surprise is structural. By the time a stock hits RSI < 30, it is almost certainly already below its rolling VWAP. There’s no independent information to add.
On NVDA over 20 years:
| Signal | Days fired |
|---|---|
| RSI < 30 alone | 152 |
| RSI < 30 AND below VWAP-10 (Q4) | 146 |
Six days across 20 years — roughly one every three years — where NVDA hit RSI < 30 while still above its 10-day VWAP. The Q4 filter is functionally identical to RSI < 30 alone. NFLX shows the same pattern: 220 RSI < 30 days, 217 in Q4 below VWAP-10.
This means VWAP is not rescuing the RSI signal on growth stocks — it’s just adding a condition that’s already satisfied 97–98% of the time. The oversold signal on NVDA is negative whether you call it “RSI < 30” or “Q4 below VWAP + RSI ≤ 30.” They’re nearly the same dataset.
Finding #2: Q4 Is the Dominant Signal — But Direction Still Splits by Stock Type
Q4 (below VWAP + RSI ≤ 30) produced 23 Bonferroni survivors — more than any other quadrant. The standout:
| Asset | Horizon | Edge (bps) | Cohen’s d |
|---|---|---|---|
| QQQ — Q4 | 1d | +112 | +0.836 |
| SPY — Q4 | 5d | +152 | +0.634 |
| SPY — Q4 | 1d | +73 | +0.620 |
| EEM — Q4 | 1d | +81 | +0.465 |
A Cohen’s d of 0.836 on QQQ at 1-day forward return is the largest effect size in all three studies in this series — larger than anything in the 856-test ETF benchmark or the individual stock study. When QQQ is below its rolling VWAP and RSI hits oversold, the next day’s return is systematically exceptional.
But the same quadrant on growth stocks tells the opposite story:
| Asset | Horizon | Edge (bps) | Cohen’s d |
|---|---|---|---|
| NVDA — Q4 | 20d | −656 | −0.474 |
| NFLX — Q4 | 20d | −639 | −0.422 |
| NFLX — Q4 | 10d | −341 | −0.320 |
The same quadrant. The same VWAP and RSI conditions. Opposite outcomes. As established in Part 2 of this series, the determinant isn’t the signal — it’s whether the stock’s price is driven by stable fundamentals (mean-reverts) or by growth expectations (trends until the narrative breaks).
Finding #3: Q3 Doesn’t Work as a Bear Continuation Signal on Defensive Stocks
The hypothesis predicts Q3 (below VWAP + RSI > 30) should produce negative returns — the stock is in a downtrend and hasn’t yet become oversold enough to bounce. The data contradicts this for defensive stocks.
Q3 Bonferroni survivors — all with positive forward returns:
| Asset | VWAP Window | Horizon | Edge (bps) | Cohen’s d |
|---|---|---|---|---|
| JNJ | 50-day | 20d | +78 | +0.185 |
| COST | 50-day | 20d | +90 | +0.160 |
| QQQ | 50-day | 20d | +79 | +0.146 |
| WMT | 50-day | 20d | +66 | +0.134 |
When JNJ, Costco, Walmart, or QQQ drift below their 50-day VWAP with RSI still in the 30–70 range, the next 20 days are positive — not negative as the hypothesis predicts. These stocks don’t wait until RSI oversold to bounce; they begin recovering as soon as the mild dislocation is established.
NVDA in the same quadrant still falls:
| Asset | VWAP Window | Horizon | Edge (bps) | Cohen’s d |
|---|---|---|---|---|
| NVDA — Q3 | 50-day | 5d | −92 | −0.135 |
| NVDA — Q3 | 50-day | 10d | −116 | −0.120 |
Below VWAP with RSI still above 30 on NVDA means the slide is just beginning. The same regime filter produces continuation on momentum stocks and reversal on defensive stocks — confirming that stock character overrides the technical regime.
Finding #4: The Biggest Surprise — Overbought Growth Stocks Keep Going
Q2 (above VWAP + RSI ≥ 70) produced one Bonferroni survivor, and it’s counterintuitive:
NVDA — Q2 above VWAP-20, RSI ≥ 70 — 20-day forward return: +210 bps, Cohen’s d = +0.152
When NVDA is overbought by RSI AND trading above its 20-day VWAP, the next 20 days are positive. This fires 615 times in 20 years — 138 independent entries — and survives Bonferroni correction.
This is momentum continuation at work. The most reliable buy signal on a secular growth compounder isn’t “oversold and beaten down” — it’s “overbought but still in trend.” The RSI > 70 reading on NVDA, when occurring above VWAP, means the trend is intact and the stock is likely to continue higher.
This inverts the conventional RSI teaching entirely. On SPY, RSI > 70 shows no edge. On NVDA in an uptrend, it’s a Bonferroni-surviving bullish signal.
Finding #5: The 50-Day VWAP Is the Only Window That Matters
| VWAP Window | Bonferroni Survivors | Interpretation |
|---|---|---|
| 10-day | 8 | Noisy — too responsive to short-term swings |
| 20-day | 10 | Standard — moderate utility |
| 50-day | 20 | Best discriminator |
The 50-day VWAP captures the meaningful trend context — it tells you whether the stock has been accumulating or distributing over the past two trading months. The 10-day VWAP is just capturing recent noise. If you’re going to use rolling VWAP as a regime filter, 50-day is the only window the data supports.
The Q1 Hypothesis: Weakly Supported
Q1 (above VWAP + RSI < 70) — the core bull regime premise — produced only 5 Bonferroni survivors, with a mean bps across all assets and horizons of just +4.8 bps.
One notable exception: NFLX Q1 above VWAP-50 at 20-day horizon: +204 bps, d = +0.134, Bonferroni significant. When NFLX is above its 50-day VWAP with RSI below 70, the 20-day forward return is meaningfully positive. NFLX’s boom periods — above long-term VWAP, not yet exhausted — do tend to persist.
TLT Q1 above VWAP-50 at 20-day: +57 bps, d = +0.152. Bond uptrends have some directional persistence.
But these are narrow exceptions. The general Q1 regime — “above VWAP, RSI has room” — does not produce the systematic upward bias the hypothesis implies across most assets.
The Quadrant Summary
| Quadrant | Survivors | Mean bps (all assets, 20d) | What it means |
|---|---|---|---|
| Q4 — below VWAP + oversold | 23 | Mixed (huge bounce for ETFs, huge drop for growth) | Strongest signal, but direction depends entirely on stock type |
| Q3 — below VWAP + RSI 30–70 | 9 | +7.4 (positive — mean-reverts for defensives) | Hypothesis predicts negative; defensives actually recover |
| Q1 — above VWAP + RSI < 70 | 5 | +13.1 | Weakly positive but not robust |
| Q2 — above VWAP + overbought | 1 | −33.5 (mostly negative — except NVDA) | Overbought in uptrend is mixed; NVDA is the exception |
The Revised Decision Framework
The data supports a more precise version of the VWAP + RSI framework — not “above goes up, below goes down,” but:
For ETFs and Defensive Stocks (SPY, QQQ, JNJ, COST, WMT):
- Best signal: Q4 — below VWAP + RSI ≤ 30. This is the strongest quantitative signal in this entire research series (d = 0.836 on QQQ at 1-day). Buy the extreme oversold reading.
- Q3 is also constructive — below 50-day VWAP with RSI 30–70 still produces positive 20-day returns. You don’t need to wait for RSI < 30 on defensive stocks.
- Q1 and Q2 show no consistent edge — don’t use VWAP as an uptrend momentum signal on these names.
For Growth/Momentum Stocks (NVDA, NFLX):
- Avoid Q4 — below VWAP + RSI ≤ 30 is a falling knife signal (d = −0.47 on NVDA). This is the most dangerous trade you can make on a momentum stock.
- Q2 on NVDA is actionable — overbought (RSI ≥ 70) above the 20-day VWAP is a Bonferroni-surviving bullish signal (+210 bps, 20d). Buy strength in trend, not weakness.
- Q3 is continuation — below VWAP on NVDA/NFLX with RSI 30–70 means the move is still developing downward. Do not fade it.
For Financials (BAC):
- No statistically robust signals in any quadrant across this study or the previous one. Macro and rate dynamics override technical indicators entirely.
Why the VWAP Framework Partially Works (and Mostly Doesn’t)
The framework’s intuition is sound for one specific case: identifying extreme oversold conditions in mean-reverting assets. When SPY or QQQ is below any rolling VWAP and RSI hits 30 — indicating both price-level displacement and momentum exhaustion — the bounce is real, large, and statistically unambiguous.
Where the framework breaks down is in its directionality claim. “Below VWAP goes down” is only true for momentum stocks where the fundamental narrative is breaking. For the majority of the market — diversified ETFs, consumer staples, healthcare — below VWAP is a buying opportunity at almost any RSI level, not a short signal.
The underlying truth is the same finding from Part 1 and Part 2 of this series: the security type determines whether “dislocated from average” means buying opportunity or trend confirmation. VWAP is a sophisticated way to measure dislocation — but it can’t change the fundamental dynamics of the underlying asset.
What to Actually Use
If you’re building a technical trading framework from these three studies, here’s what the data supports:
| Setup | Signal | Action | Best on |
|---|---|---|---|
| Below 50-day VWAP + RSI ≤ 30 | Q4 | Long, 1–5 day hold | SPY, QQQ, EEM |
| Below 50-day VWAP + RSI 30–70 | Q3 | Long, 20-day hold | JNJ, COST, WMT |
| Above 20-day VWAP + RSI ≥ 70 | Q2 | Long, 20-day hold | NVDA specifically |
| Above 50-day VWAP + RSI < 70 | Q1 | Long, 20-day hold | NFLX (narrow) |
| Below VWAP + RSI ≤ 30 | Q4 | Avoid or short | NVDA, NFLX |
| Below VWAP + RSI 30–70 | Q3 | Avoid or short | NVDA |
Limitations
VWAP is a smoothed price-volume average, not a causal mechanism. Its statistical power in this study is primarily as a regime classifier for when RSI oversold signals are reliable. It does not create the edges — it identifies which asset class context you’re in.
50-day VWAP uses 50 trading days of history — the warm-up period and window selection affect signal count. Tests with fewer than 10 signal observations were excluded; some asset-window combinations have limited data depth in the early 2004–2006 period.
Serial correlation: Regime days are not independent observations. A Q1 period lasting 60 consecutive days contributes 60 data points but represents a single regime entry. We report both n_days and n_entries (independent regime transitions) for this reason. All QQQ Q4 results showing d = 0.836 are based on 38 independent regime entries across 20 years — meaningful but not a high-frequency signal.
Pre-cost: All edges are before transaction costs. The 1-day QQQ Q4 signal fires ~2 times per year on average. At any reasonable cost structure this remains actionable. Higher-frequency signals require cost adjustment.
Conclusion
The VWAP + RSI framework as commonly taught — above VWAP is bullish until overbought, below VWAP is bearish until oversold — is not statistically supported in its general form. VWAP adds almost no incremental filtering power to RSI alone because oversold readings already imply below-VWAP status on most assets.
What the study does confirm is a more precise version of the framework:
- The strongest quantitative signal in this entire research series (d = 0.836) is QQQ below VWAP with RSI ≤ 30 at the 1-day horizon
- Defensive stocks mean-revert even in Q3 — you don’t need RSI < 30 to get a recovery on JNJ or COST
- Growth stocks require the opposite playbook — buy overbought in trend (NVDA Q2), not oversold below VWAP (NVDA Q4)
- The 50-day VWAP is the only window with robust discriminating power; the 10 and 20-day windows add noise
The underlying principle remains constant across all three studies: the asset’s fundamental character — mean-reverting or trend-following — determines how technical signals behave. No regime filter overrides it.
This is Part 3 of the Reigraph Research Technical Indicator Series. Part 1: Do Technical Indicators Actually Work? 856 Tests Part 2: Same Signal, Opposite Result — Growth vs. Defensive Stocks
Reigraph Research · May 2026
Not investment advice. All results are pre-transaction-cost based on historical data. Past statistical relationships do not guarantee future performance.