VWAP Trading Strategy Backtest: Does the QQQ Result Generalise?

1. What the original VWAP study tested

Volume Weighted Average Price (VWAP): The Holy Grail for Day Trading Systems (Carlo Zarattini and Andrew Aziz, SSRN 4631351, 13 November 2023) tests a rule on QQQ using 1-minute candles from 2 January 2018 to 28 September 2023: long while QQQ trades above VWAP, short while below, flat overnight. It reports +671% against +126% for buying and holding QQQ, with a largest fall of 9% against 36%, and describes the work as exploratory.

We rebuilt the long side of that strategy and asked three questions: does the paper's pattern appear in our rebuild on QQQ over the paper's years; does the advantage persist in the years after and on other stocks; and does it survive trading costs. How VWAP and Anchored VWAP work explains the indicator and the rule with diagrams.

2. The VWAP rule we rebuilt

Two references are used. Buying and holding buys at the close of the window's first candle and sells at its last close. Holding each session buys at 09:35 and sells at 16:00 every day. The second is the like-for-like comparison for an intraday rule: much of QQQ's rise in these years happened overnight, between one day's close and the next day's open, which a rule that is flat every night cannot earn.

This is a test of one VWAP rule. A different candle length, signal, execution, anchor or exit makes a different strategy, which needs its own test. The rule is built in GU Analyser from Anchored VWAP anchored at the market open and run through the app's own backtester.

3. Important differences from the original test

Execution is the most important difference. The signal is known only once the candle has closed, but the backtest fills the order at that same closing price. A live order would fill later, at a price such as the next candle's open. Because the rule's gross advantage is small relative to how often it trades, this difference could change the result. The original paper entered at the start of the following candle, so the reconstruction approximates its test.

• Candles and volume. 5-minute candles from one US exchange, IEX, whose volume is a few per cent of all US trading in these stocks. VWAP is volume-weighted, so the source of the volume moves the line; the paper used consolidated 1-minute data. Study 007 shows how the price feed can move an intraday backtest.
• Window. QQQ from 1 May 2018 to 25 September 2026: our data carries no traded volume before May 2018, so the paper's first four months cannot be tested. 18 half-day sessions and 4 sessions with almost no volume are left out.
• The stocks. 149 large US stocks and SPY, 1 April 2026 to 25 September 2026, from a list with selection biases described in section 6.
• Specification. The main test specification was fixed in GU Analyser's internal record on 4 October 2026, before the results were computed: the rules, windows, screens, references, costs and the decision rule. That reduces the risk of changing the test after seeing the result. The record is internal to GU Analyser, and the limitations of the data and of the stock list still apply. The comparison with the paper's own figures in section 4 was added while the run was in progress, before any result existed.
• Checks. The app's VWAP line equals an independent calculation from the same candles on every bar checked; every trade's fill was checked against its candle; no position was held overnight.

Exploratory: the rule with orders at the next candle’s open

We added this check after the results were known, at the founder's request, to measure the effect of the same-close fill. It is exploratory: it was not part of the test fixed in advance. Both orders fill at the open of the candle after the signal; the last exit signal is the 15:50 candle, filled at 15:55's open, so nothing is held overnight. The rule and the references are otherwise unchanged, and every trade was checked to fill at a candle's open on the same day.

With next-candle fills, QQQ over the paper's years returned +111.2%, against +141.9% with same-close fills, behind buying and holding at +122.9%; its daily comparison with holding each session was +0.019% a day, no clear difference. On QQQ in the paper's years, the same-close fill flattered the rule. In the later QQQ years, SPY and the median stock, next-candle fills did slightly better than same-close fills, and the rule stayed far behind buying and holding. On the stocks it beat buying and holding on 48 of 149. At 0.02% per side the account fell below its starting value on QQQ in both periods with either fill, and on 130 of the stocks with next-candle fills. The rest of this study reports the same-close version, which is the specification fixed in advance.

4. QQQ during the original study period

Our QQQ prices match the paper's opening and closing prices to within $0.02. Over May 2018 to September 2023, QQQ rose $238 a share across the candles that followed a close above VWAP and fell $137 across those that followed a close below it. The direction matches the paper's +$320 and −$280; the size is about 74% and 49% of theirs. The lower figures are consistent with the shorter window (QQQ moved +$4 a share in the four missing months), 5-minute candles where the paper used 1-minute, and a different volume source; this test does not isolate how much each contributes.

At the paper's near-zero cost, the long side's compounded return was higher than buying and holding over these years, +141.9% against +122.9%, with a largest fall of 11% against 36%. The pre-specified daily comparison did not find a clear difference from holding during market hours: the estimated advantage was +0.030% a day, and its interval includes zero. A higher final balance does not by itself show a reliable edge; the daily comparison asks whether the difference is larger than ordinary variation in the sample. The paper reports 36% for buying and holding QQQ's largest fall, the same figure as ours.

5. QQQ after the study

From 29 September 2023 to 25 September 2026, with the rule unchanged, it returned +9.7%; holding each session returned +12.0% and buying and holding +105.9%. Its falls stayed smaller (8% against 23%), and it did not keep its lead over buying and holding.

6. The strategy on 149 stocks

At the paper's near-zero cost, the rule beat buying and holding on 46 of 149 stocks and beat holding each session on 69. The median stock returned +0.0% with the rule, -0.2% holding each session and +11.2% held throughout. On SPY, the S&P 500 fund and the closest relative of QQQ here, the rule returned +0.9% against +6.1% and +19.2%. SPY gained +19.2% over the window, and a rule that is out of the market for part of each day gives up part of such a rise.

7. The pre-specified comparison

The comparison fixed in advance sets the rule against holding the same stock during the same trading hours, day by day, and asks whether the difference is large enough to separate from ordinary day-to-day variation.

At the paper's cost there was no clear difference from holding each session in any of the four tests. At 0.02% per side the rule was clearly behind holding each session in all four.

Method. The 95% intervals come from a block bootstrap of consecutive trading days, 2,000 resamples, at three block lengths (20, 5, 60 days for QQQ; 5, 2, 10 for the stocks, resampled together). We treated a difference as clear only when all three 95% bootstrap intervals stayed on the same side of zero.

8. What happens after trading costs?

The rule made about 3.9 round trips a day on QQQ, each a buy and a sale, so each assumed cost is paid about 8 times a day. At 0.02% per side it finished every calendar year from 2018 to 2026 below where the year started on QQQ. At 0.02% per side the account fell below its starting value on 127 of the 149 stocks.

A market pattern and a profitable trading strategy are different things: a signal can carry information about future prices and still fail to make money after execution, timing and the cost of repeated trades. What happens when trading costs are added? Study 011 measures the cost at which the rule's lead disappears.

9. What this backtest cannot tell you

These results describe past trading under stated assumptions. They leave these questions open:

• The short side. The paper's +671% includes it.
• The paper’s 1-minute implementation or consolidated volume. The paper's data reads each cross sooner and weights VWAP by all US trading.
• Real fills. The specified test fills at the signal candle's close and does not model the delay between a signal and an order, the bid-ask spread or market impact; section 3 measures one alternative.
• Other VWAP rules. A different candle length, signal, anchor or exit is a different strategy.
• The first months of 2018. Our data has no traded volume from January to April 2018.
• US stocks in general. Six months, one rising market, a list with selection biases.
• Dividends. Returns are price only, as in the paper.

10. Conclusion

The rebuild finds the same direction as the 2023 QQQ study: QQQ gained more on the candles following closes above VWAP than it lost on those following closes below it, but the effect was smaller. The 5-minute long-only rule also finished ahead of buying and holding QQQ over the study's period at near-zero cost, though the daily comparison fixed in advance found no clear difference from holding during market hours.

The advantage was not reproduced in the later QQQ period: +9.7% against +105.9% from September 2023 to September 2026. In the six-month stock test the rule beat buying and holding on 46 of 149 stocks. At 0.02% per side the account fell below its starting value on QQQ in both periods and on 127 of the stocks.

In the check added after the results, filling each order at the next candle's open put the rule behind buying and holding QQQ in the paper's own years (+111.2% against +122.9%).

The rebuild supports the existence of the historical QQQ pattern. A robust trading edge would have to hold in later years, on other stocks and after trading costs, and in these tests the rule met none of the three.

Sources

• Carlo Zarattini and Andrew Aziz, Volume Weighted Average Price (VWAP): The Holy Grail for Day Trading Systems, SSRN 4631351 (13 November 2023)

The paper's figures are as its authors reported them.