LAA (Lethargic Asset Allocation)

Data it needs

Monthly prices for QQQ, IWD, GLD, IEF, SHY and SPY, plus the UNRATE series from FRED. The signal only changes when the two conditions coincide, so there is no need to watch it daily.

At the close of the last business day of each month:

1. Base portfolio. 25% goes to IWD, 25% to GLD and 25% to IEF. These three are always held.
2. Unemployment trend. Compare the latest unemployment rate with the average of the last 12 monthly readings (including the latest). If the ratio minus 1 is positive, unemployment is rising.
3. S&P 500 trend. It is considered bearish if SPY's price is below its 10-month simple moving average.
4. The fourth 25%. If both conditions hold at the same time, that 25% goes to SHY (1-3 year Treasury bonds) instead of QQQ. In any other case, it stays in QQQ.

With only one of the two conditions, nothing happens. That is what makes the portfolio lethargic: the labour market has to deteriorate and the stock market has to already be falling.

Two decisions made by this site

The unemployment data arrives a month late. The paper's main backtest uses the unemployment figure for the same month as the signal, but that figure is not known at the month's close: it is published well into the next. We use the latest published figure, that is, the previous month's. It is the variant the paper itself tests in its Fig. 17, with almost identical results (10.4% a year against 10.5% over 1949-2019). That way the signal does not use information that did not exist on the calculation date.

IWD is the Russell 1000 Value. The paper's introduction labels IWD as growth and IWF as value, the opposite of what they are. The abstract and Fig. 21 use IWD as the Russell 1000 Value, and that is what we take.

How this month looks

With the closes of September 30, 2026, the two conditions do not hold at the same time, so the signal keeps 25% in QQQ. The "Current signal" table above shows the complete portfolio.

With real ETFs from Jan 2005 to Sep 2026, our calculation gives an annual return of 9.8%, a maximum monthly drawdown of -18.9% and volatility of 9.0%. The S&P 500 over the same long period: 10.8% a year and a maximum drawdown of -51%. Costs are 0.1% per trade. Average annual turnover is 31%. Since on average only 2.3% of the portfolio has been in SHY instead of QQQ, a good part of that turnover does not come from signal changes but from the backtest returning to 25% weights every month.

And since it was published?

Since Dec 2019, the paper's publication date, LAA returns 10.8% a year with a maximum drawdown of -18.9%. It is a short period and proves nothing on its own, but it is a stretch that no adjustment by the author could have taken into account.

What the paper publishes

The author presents figures from 1949 to October 2019, with approximate series from before ETFs: 10.5% a year, volatility of 8.5% and a maximum drawdown of 15.0% (Fig. 21 of the paper). A 60/40 portfolio of SPY and IEF gives in the same paper 9.5% a year and a drawdown of 29.5%. He also separates an "in-sample" stretch (February 1949 to June 1981: 10.6% a year) and an "out-of-sample" one (June 1981 to October 2019: 10.4%). Those figures are the author's and we have not recalculated them: our ETF data only starts in Jan 2005, and Keller's, before ETFs, are proxies that could not be bought.

• Few signal changes: three of the four assets never change and the fourth does so only when two conditions coincide. If you rebalance weights only when the signal changes or once a year, as the author suggests, you sell much less than with strategies that rotate every month.
• Simple diversification: tech stocks, "value" stocks, gold and bonds. It does not depend on a single asset.
• A portfolio chosen with hindsight: the author himself acknowledges that the four assets were chosen by optimising a return-and-risk indicator over the in-sample stretch, and that QQQ takes advantage of the later tech boom. Nobody guarantees that the same split will work in future.
• Pre-ETF series: before 1970 QQQ is replaced by the Fama-French technology sector, and all history before each ETF consists of proxies built by the author, not tradable prices.
• Macro data with delay and revisions: the main backtest uses unemployment without later revisions. The real figure arrives late and is corrected afterwards; the paper only covers the one-month delay in Fig. 17.
• External criticism: Newfound Research has published criticism of unemployment signals with a 12-month average and of SPY with a 10-month average. The author himself cites them; we have not consulted them.
• Errata in the paper: besides the inverted IWD and IWF labels, an in-sample period is misprinted on p. 2 (it says 2019 where it should say 1981). Our observation.

1. At the start of the month check this page's signal: it shows whether the fourth 25% is in QQQ or SHY.
2. If it has not changed, there is nothing to do.
3. If it has changed, sell one asset and buy the other. Use the rebalancing calculator to work out the shares.
4. Map each ETF to the fund you actually hold.
5. Keep a record of your trades for your tax return.

For practical use the author suggests rebalancing to equal weights only when the signal changes or once a year. Our backtest recalculates the weights every month, and that monthly rebalancing is not in the paper's costs.