The anchored VWAP strategy works like this. You pick a starting point on the chart, for example the last major low. From there the indicator draws a line with the average price every buyer has paid since. While Bitcoin is above the line, you are in. When price falls below it, you sell.
On a chart this often looks convincing. Price drops, touches the line and turns. Twice, three times. It looks like support.
Before you build a decision on it, ask one question. Is the line there because the market respects it, or because you put it exactly there?
The promise: the average price of everyone who bought since the anchor
A normal moving average always looks back the same distance, say 200 days. An anchored VWAP starts at a point you choose: a low, a high, a news event. From there it computes the average price of every trade, weighted by volume.
The idea is appealing. If price sits above the line, everyone who bought since the anchor is in profit on average. If price falls back to it, they defend their entry.
Brian Shannon made the tool popular, among other things with his book "Maximum Trading Gains With Anchored VWAP" (2023). For Bitcoin, the TradingView author AstralVision published an indicator on 8 May 2026 that anchors at every year start since 2013. Its description says anchored VWAPs are "among the most reliable dynamic support and resistance levels available".
And on Bitcoin it works, at first sight. We tested a simple rule: long while the daily close is above the line, out as soon as it falls below. The line starts at the last confirmed turning point (a high or low that stands out over 40 days on each side). From July 2020 to October 2026 that returned 44.8 % a year after fees. Holding returned 42.7 %. The worst drawdown was −61.0 % instead of −76.6 %. The rule also passes the platform's out-of-sample test.
If the story ended here, AstralVision would be right. But one number is still missing: what would have happened if the anchor had been somewhere else?
The test: the same rule with a random anchor
With an anchored VWAP, the anchor is the one thing you decide. So that is what we tested.
For five common anchor rules, the original ran against 1,000 copies. Each copy re-anchors at the same moments as the original, but at a random point within the previous 252 trading days. Everything else stays the same: long above the line, out below, 0.1 % fee per side.
If the anchor carries information, the original rule has to beat most of its random copies.
| Anchor rule | Annual return | Random anchors that did better |
|---|---|---|
| Lowest low of the last 252 days | 29.9 % | 99 % |
| Highest high of the last 252 days | 33.6 % | 98 % |
| Last turning point, confirmed over 10 days | 36.4 % | 87 % |
| Last turning point, confirmed over 20 days | 45.4 % | 62 % |
| Last turning point, confirmed over 40 days | 50.1 % | 39 % |
BTCUSDT, daily candles, 1 July 2020 to 6 October 2026, execution at the close. Holding: 42.7 % a year. With execution only at the next day's close, the 40-day rule lands at the 44.8 % from above.
None of the five rules places the anchor better than chance. The best one sits in the middle of its own random copies, the others below. So the rule from the opening, with its 44.8 %, did not work because of its anchor. It worked because Bitcoin went up, and a long-above-the-line rule is long most of the time in a rising market, wherever the line is drawn.
The anchor is a choice, not information.
Which leaves the question of why the most popular starting point does worst of all.
The 52-week low: a line that sits below price by design
Anchoring at the yearly low sounds like the smartest anchor there is. The line shows the entry price of everyone who bought the bottom.
That is exactly the problem. An average that starts at the lowest point almost always sits below price. So the rule is almost always long, on 79 % of days on average. When the market makes a new low, the anchor jumps there, the line restarts from below, and the rule buys again.
| 52-week low as anchor | Holding | |
|---|---|---|
| 2021 | −3.1 % | +59.8 % |
| 2022 | −52.0 % | −64.2 % |
| Worst drawdown | −78.9 % | −76.6 % |
In spring 2021 Bitcoin fell almost by half from its April high. The rule stayed long the whole way, because the line sat far below price, and only exited on 23 May at $34,655. It then bought four small rebounds and lost on every one. In 2022 it was invested on only 35 % of days and still lost about half: 25 trades, 22 of them losers. Every new low reset the anchor, and every small rebound was a buy signal. From April 2021 to January 2023 that adds up to −78.9 %, deeper than simply holding.
A line that always sits below price cannot warn you about a crash.
If that is so clear, why do the charts look so convincing?
Why the chart still convinces
Because you get to pick it. Move the anchor a few candles and the line moves. Over six years there are thousands of possible starting points, and in hindsight there is always one where price bounces neatly three times.
The random copies show how far apart the outcomes are. The same rule with a different anchor lands anywhere between 17 % and 79 % a year. Holding sits at 42.7 %. A chart with a clean bounce mostly shows you which anchor someone chose.
Passing a test does not protect you from this either. The 40-day rule passes the platform's out-of-sample test: execution only at the next day's close, net of fees, the last 30 % of the trades unseen. Still, 39 % of the random anchors did better. An out-of-sample test checks whether a rule keeps working on new data. It does not check whether its anchor knows anything.
A passed test proves the rule runs, not that the anchor matters.
Then surely the volume adds something? It is the V in the name.
The objection: "But the volume makes the difference"
We checked. Every rule ran twice: once as a true VWAP, once as a plain average of prices since the anchor, without volume weighting.
| Anchor rule | VWAP minus unweighted average |
|---|---|
| Lowest low, 252 days | +6.1 percentage points |
| Highest high, 252 days | −4.5 |
| Turning point, 10 days | +1.6 |
| Turning point, 20 days | −1.6 |
| Turning point, 40 days | −2.9 |
Volume helps twice and hurts three times. Use the closing price instead of the average of high, low and close, and two of the signs flip. There is no consistent direction.
Volume is neither the problem nor the fix. The anchor is the problem.
What to do with this
If you use anchored VWAPs:
- Don't read a bounce as proof. First ask who set the anchor and whether another spot would have looked just as good.
- Don't use the 52-week low as your exit line. It sits below price by design and in May 2021 only got you out after the crash.
- Test your own anchor rule against random anchors before you trust it with money. If it does not clearly beat more than 95 % of its random copies, your anchor carries no information.
For comparison, here is how the anchor rules do against well-known strategies from the backtester, same window, after fees.
| Strategy | Annual return | Worst drawdown | Closed trades |
|---|---|---|---|
| Daily candles, 1 Jul 2020 to 6 Oct 2026 | |||
| Holding | 42.7 % | −76.6 % | – |
| EMA Cross | 45.7 % | −60.4 % | 45 |
| Anchored VWAP, 40-day turning point | 44.8 % | −61.0 % | 46 |
| 200-day line | 41.8 % | −63.4 % | 26 |
| Golden Cross | 36.2 % | −56.7 % | 6 |
| Wuguan Master | 34.6 % | −48.5 % | 35 |
| Supertrend | 26.7 % | −53.1 % | 28 |
| Anchored VWAP, 52-week low | 24.9 % | −75.2 % | 64 |
| RSI/SMA Cross | 24.7 % | −59.5 % | 182 |
| Weekly candles, 6 Jul 2020 to 5 Oct 2026 | |||
| Holding | 42.1 % | −75.2 % | – |
| WMA Trend | 43.1 % | −63.8 % | 3 |
| RSI/SMA Cross | 42.0 % | −40.4 % | 22 |
| Bull Market Support Band | 40.7 % | −46.9 % | 12 |
| RSI oversold/overbought | 12.7 % | −33.0 % | 1 |
BTCUSDT, default settings, execution at the close of the next candle, 0.1 % fee per side. Strategies: arena_compare_strategies and arena_run_backtest; anchored VWAP: validate_strategy with our own signal list.
The best anchor rule lands in the middle of the simple trend filters, the worst at the bottom. No strategy in the table beats holding by more than 3 points, and some differences rest on a handful of trades. The gap between them is smaller than the 17 % to 79 % spread that the starting point alone produces for an anchored VWAP.
An anchored VWAP is a trend filter with one extra dial, and the dial adds nothing.
If all you want is a trend filter, use one without a chosen starting point, such as the 200-day line ("Index Regime (200-Day Trend)" in the backtester). You can check your own signal lists with the validation tool.
What is not claimed here: that anchored VWAPs work nowhere. The test covers Bitcoin on daily candles, long only, five anchor rules. Intraday equities and anchors at fixed dates such as earnings releases are a different question. The open measurement is exactly that: is there a market where an anchor rule beats more than 95 % of its random copies?
How it was measured
Data: Binance BTCUSDT, daily candles, 1 July 2020 to 6 October 2026, 2,289 days. VWAP from the average of high, low and close, weighted by daily volume. Rule: long when the close is above the line, otherwise cash; return from that close, 0.1 % fee per side. Turning points only count as known once the confirmation days have passed, so 10, 20 or 40 days later. Using them immediately means looking into the future.
Random anchors: 1,000 per rule, re-set at the same moments as the original rule, uniformly distributed over the 252 days before. Cross-checks with 200 anchors, a different random seed, and the close instead of the average price change none of the findings. The rules were fixed before the random copies ran.
Returns in the tables use execution at the close. The 44.8 % for the 40-day rule comes from the platform's validation tool with execution at the next day's close (46 closed trades, 70/30 split). The worst drawdown of −61.0 % uses the same execution, marked day by day, just like holding. Holding starts on the same day as the rule in every comparison, 1 July 2020, not at the respective anchor.
FAQ
What is a "confirmed turning point"? A low that is lower than the 40 days before it and the 40 days after it (highs accordingly). You only know it 40 days after the low. A backtest that anchors at the low itself uses knowledge that did not exist at the time.
Why do some random anchors beat holding? Because the rule is often out during steep crashes. That is a property of the "long above a line" rule, not of the anchor. The 200-day line has the same property without you having to choose a starting point.
Does this apply to intraday VWAP? The classic VWAP restarts at every session open. There is no freely chosen anchor. Only the anchored version on daily candles is tested here.
Not investment advice, not a recommendation, not a forecast — historical patterns are no promise.
Study the Past — Improve your Future 🥋