VAA-G12 (Vigilant Asset Allocation)
Twelve risk assets and three cash assets. The best two are chosen, but it only takes four of the twelve being negative to move the whole portfolio to Treasury bonds. It spends more than half its time out of risk.
What VAA-G12 is
VAA stands for Vigilant Asset Allocation. It was published by Wouter Keller and Jan Willem Keuning in July 2017 in the paper Breadth Momentum and Vigilant Asset Allocation (VAA): Winning More by Losing Less (SSRN 3002624). "G12" is the global universe of 12 risk assets. There is also a 4-asset version, VAA-G4, with its own page.
The key phrase is breadth momentum: instead of looking at a canary or at the chosen asset, it counts how many of the 12 risk assets have negative momentum. The more there are, the more money moves to cash. Protection is gradual rather than all or nothing, although with the default parameters there are only three steps.
Quick facts
- Author
- Wouter J. Keller and Jan Willem Keuning
- Year
- 2017
- Origin
- Paper
- Access
- Public and free
- Type
- Tactical
- Assets
- 14
- Frequency
- Monthly
- Difficulty
- Medium
Who it is for
Anyone who wants to study how you go from a per-asset filter to a market-breadth one and accepts a portfolio concentrated in two positions and heavy turnover.
Who it is not for
Anyone looking for few trades a year, tax efficiency or a portfolio that is diversified at every moment.
Who it is for and who it is not for
It works as an example of a breadth-momentum strategy, with complete and public rules. It asks you to accept high turnover and a portfolio of few positions, and suits those who can trade once a month without doubting every signal.
Current signal
VAA-G12 (Vigilant Asset Allocation)
Calculated with the closes of September 30, 2026.
What changes from last month
Rules, step by step
At the close of the last business day of each month:
- 13612W momentum of the 12 risk assets and the 3 cash assets: 12 times the 1-month return, plus 4 times the 3-month return, plus 2 times the 6-month return, plus the 12-month return. The paper averages it; for the sign and for ranking assets it makes no difference. It weights the latest month heavily, so it reacts fast.
- b = number of risk assets with 13612W less than or equal to zero.
- Cash fraction with the parameters T = 2 and B = 4: the one given by the table below the list.
- Rank the risk assets by 13612W and take the T = 2 best, at 50% each. The part going to cash replaces the worst of that pair ("easy trading": with 50% in cash only the best risk asset remains at 50%; with 100% none remains).
- The cash part goes entirely to the best of SHY, IEF and LQD by their 13612W, even if that momentum is negative. There is no absolute filter on cash.
- Hold until the close of the following month.
| Risk assets negative (b) | Cash fraction |
|---|---|
| 0 or 1 | 0% |
| 2 or 3 | 50% |
| 4 or more | 100% |
The ETFs used by the engine are SPY, IWM, QQQ, VGK, EWJ, VWO, VNQ, GSG, GLD, TLT, LQD and HYG as risk assets, and SHY, IEF and LQD as cash. LQD appears in both lists.
Decisions made by this site
- Cash with SHY, not BIL. The paper's conclusions (p. 35) cite BIL as the cash asset, but the method and the backtests use SHY (1-3 year Treasury bonds). We follow the method.
- VWO and VNQ instead of EEM and IYR. The paper's text names EEM and IYR; note 11 clarifies that in nearly all backtests proxies for VWO, VNQ, VEA and BND were used.
- Commodities with GSG. It is the ETF for the S&P GSCI index; DBC, which tracks another commodity index, would be the alternative.
- Tie-breaks. The paper does not say what to do if two assets have exactly the same 13612W. The engine applies no rule of its own; with real prices an exact tie is very unlikely.
- 0.1% costs per trade, like the paper (p. 11).
It is a concentrating strategy: with 50% in cash, the portfolio is 50% in a single risk asset and 50% in a single cash asset.
Backtest and out-of-sample period
With real ETFs from Jun 2008 to Sep 2026, VAA-G12 returns 7.1% a year, with volatility of 10.0% and a maximum monthly drawdown of -29.6%. On average 55% of the portfolio is in cash and it turns over at an annual rate of 770%.
Since the paper's publication (Jul 2017) it returns 4.8% a year, with a maximum drawdown of -29.6% and a Sharpe of 0.27.
What the authors publish
Figures from the paper's tables, with 0.1% costs and calculated with ETF proxies built by the authors since 1970:
| Series | Period | Annual return | Volatility | Worst drawdown | Sharpe | Source |
|---|---|---|---|---|---|---|
| VAA-G12, full sample | 1970-12 to 2016-12 | 15.6% | 10.2% | -13.0% | 0.85 | Table 4, p. 16 |
| VAA-G12, out of sample | 1993-12 to 2016-12 | 10.5% | 10.3% | -13.0% | 0.51 | Table 3, p. 16 |
| VAA-G12, optimised stretch | 1970-12 to 1993-12 | 21.0% | 10.0% | -5.8% | 1.24 | Table 2 |
| 60/40 SPY/IEF (paper's reference) | 1970-12 to 2016-12 | 9.6% | 9.7% | -29.4% | 0.28 | Table 4, p. 16 |
The paper's Sharpe is calculated against the return of a cash portfolio (EWC), not against the risk-free rate, so it is not comparable with ours.
Two clarifications for reading the two columns. "Out of sample" has the paper's meaning here: the parameters were optimised over 1970-1993 and validated over 1993-2016. That is not the same as "since publication", which is the stretch our engine measures. And the authors' figures and ours are not from the same period or the same data, so the comparison only serves as an order of magnitude. The maximum drawdown of our backtest (-29.6%) against the paper's 13.0% is a difference we have not broken down. Among other things, our history only uses real ETFs (with 2008-2009 and 2022 inside it) while the authors' is almost entirely proxies they built themselves.
4.8% a year · max drawdown -29.6% · S&P 500 over the same period: 15.0%
An "out-of-sample" result: the period after the rule was published, which could not have been used to tune the rules.
Growth of $100 (logarithmic scale) since , with 0.1% costs per trade. Data: US-listed ETFs, total return (dividends reinvested), before taxes.
Drawdowns from peak
| Strategy | CAGR | Volatility | Max drawdown | Sharpe |
|---|
Year by year
| Year | VAA-G12 (Vigilant Asset Allocation) | S&P 500 | 60/40 |
|---|---|---|---|
| 2026 (to Sep 2026) | +0.8% | +12.7% | +5.8% |
| 2025 | +23.9% | +17.7% | +13.8% |
| 2024 | +2.0% | +24.9% | +14.7% |
| 2023 | +3.0% | +26.2% | +17.2% |
| 2022 | -15.3% | -18.2% | -17.0% |
| 2021 | -0.5% | +28.7% | +15.9% |
| 2020 | +24.4% | +18.3% | +15.0% |
| 2019 | +4.0% | +31.2% | +21.9% |
| 2018 | +3.8% | -4.6% | -2.4% |
| 2017 | +11.6% | +21.7% | +14.0% |
| 2016 | +2.0% | +12.0% | +7.6% |
| 2015 | -0.8% | +1.2% | +1.3% |
| 2014 | +1.8% | +13.5% | +11.7% |
| 2013 | +2.4% | +32.3% | +16.9% |
| 2012 | +17.2% | +16.0% | +11.1% |
| 2011 | +8.6% | +1.9% | +7.4% |
| 2010 | +18.1% | +15.1% | +12.8% |
| 2009 | +17.4% | +26.4% | +13.1% |
| 2008 | +14.8% | -36.8% | -14.9% |
The first year may be partial. Returns are net of trading costs.
Taxes on rotating
The results above are before taxes. This strategy re-assesses the portfolio every month: historically it turns over an average of 770% of the portfolio a year and spends 55% of months in defensive mode. Every change is a sale.
In most countries each sale of an ETF in a taxable account realises a gain or loss that may be taxed, so rotating monthly is best done inside a tax-advantaged account (such as an IRA or 401(k) in the US, or an ISA or SIPP in the UK) where one is available. Rules vary by country and by account: check yours, or ask a tax adviser. See our methodology for how the results are calculated.
Criticisms and limits
The parameters T=2 and B=4 were optimised over 1970-1993 and, out of sample, the paper's return falls from 21.0% to 10.5%. Almost its entire history uses ETF proxies built by the authors. It spends close to 59% of the time in cash on average and trades a lot (monthly turnover estimated around 60%). Our backtest with real ETFs and 0.1% costs does not reproduce the paper's very low maximum drawdown (13.0%).
- Parameter optimisation. T and B were chosen by optimising over 1970-1993. The authors themselves acknowledge the risk of data-mining bias and that is why they separate sample and validation (pp. 7-8 and Table 2). The out-of-sample result is half the optimised one: 10.5% against 21.0% (our calculation from Tables 2 and 3).
- History made of proxies. Almost the entire series in the paper are ETF proxies built by the authors and not real prices (section 7, pp. 9-10).
- A lot of time in cash and a lot of turnover. About 59% of the time in cash on average and monthly turnover that the authors estimate at close to 60% (p. 16). Each turn is a sale with tax impact.
- Very concentrated positions. Two risk assets at most, and only one when cash is 50%.
- Cash without a filter. If the best cash asset has negative momentum, the money goes there anyway. In a year like 2022, with bonds and stocks falling together, there is no full refuge.
How to run it
- Wait for the close of the last business day of the month. The signal is calculated with that close.
- Look at the current portfolio in "Current signal" and compare it with yours.
- Map each ETF in the signal to the fund you actually hold.
- Work out the orders with the rebalancing calculator.
- Trade on the first business day of the following month. The backtest assumes execution at the close.
Similar strategies
Frequently asked questions
What is 13612W momentum?
It is the weighted sum of the 1, 3, 6 and 12-month returns with weights 12, 4, 2 and 1. It gives heavy weight to the latest month, which is why it reacts fast to trend changes, and also why it generates false alarms.
What is the difference between VAA-G12 and VAA-G4?
G4 has only 4 risk assets and moves 100% to cash as soon as one of them has negative momentum. G12 has 12 and lets cash in step by step. G4 is more aggressive and concentrates everything in one asset.
Why does it use SHY and not BIL as cash?
The paper's method and its backtests use SHY (1-3 year Treasury bonds). The conclusions cite BIL while the method uses SHY. This site follows the method.
Is this investment advice?
No. It is educational information about a published strategy. Neither backtests nor past performance guarantee future results, and you can lose money.
Sources and review date
- Keller and Keuning, "Breadth Momentum and Vigilant Asset Allocation (VAA): Winning More by Losing Less", SSRN 3002624 (v0.99, July 2017) · accessed October 10, 2026
Page last reviewed on October 11, 2026; figures and signal are recalculated at every monthly close (latest: September 30, 2026). How we work: editorial policy and methodology.
Educational and informational content. It is not investment advice or a personal recommendation. The results shown are historical simulations (backtests) using US-listed ETFs; past performance does not guarantee future results and you may lose some or all of the capital invested. Real returns will depend on your costs, taxes, currency and execution timing.