Five years of returns gave higher risk forecast loss than ten years in 20 annual US market tests, 2005–2025
We tested whether five years of past returns give better monthly US market risk forecasts than ten years. At each of 20 annual origins, we used the official French Data Library archive for that year. Inputs stopped in July. Forecasts stayed fixed from October through September. The test covered October 2005–September 2025, with 240 scored months. Risk was the monthly excess-return second moment. The primary QLIKE gain, defined as ten-year loss minus five-year loss, was -0.300. Its descriptive 95% stationary bootstrap interval was -0.776 to +0.004. The planned positive-gain criterion was not met. The October 2008 block supplied about 75% of the loss disadvantage. The point gain stayed negative after any one annual block was removed and under later-vintage checks. Five-year and ten-year mean return forecasts had R2 of -1.652% and -0.009% against the all-history mean. In unlevered paper accounts with 10 bp trading costs, the five-year risk rule had 1.81 percentage points less annual utility than the ten-year rule. The interval was -4.36 to +0.33 points. No reliable return forecast or trading advantage was established. The main limits are 20 annual decisions, stale inputs, noisy squared-return targets, and assumed archive release dates. The package preserves the plan, code, source hashes, forecasts, bootstrap draws, figures, and independent checks. Raw sources are retained separately with their original rights. Twelve later paper forecasts are preserved without scored outcomes.
stock returns · risk forecasts · point-in-time data · data revisions · chronological evaluation · paper trading · negative result
References
6- usesFive published US portfolio return signals did not pass a Holm-corrected holdout test after their original samples to August 2026, and industry seasonality changed to a net loss sci:sha256:6e3392a435bad5ed24d8ecb7f11468605a41e192a217e8dfc6393ca5859be052
Its results and data-revision check motivated the use of archived releases in the main test. No code was reused.
- usesIn 2019–2026, 12-month factor momentum on 23 public equity factors gave no clear net gain over equal factor weights (gain +0.06%/month, 95% CI −0.30 to +0.44) sci:sha256:605a364268f178775b231b39cab8d4bba7e8cb3a0f6c07b7fc8934fa835b4d2e
Its negative factor-timing test and vintage limits informed the question and baseline choices. This study tests broad-market second moments.
- usesdoi:10.1093/rfs/hhm014
The historical mean is the return forecast baseline.
- doi:10.1111/jofi.12513
Risk-based allocation motivates a secondary paper simulation. The slow annual rule here is not a replication of this paper.
- usesdoi:10.1016/j.jeconom.2010.03.034
This work motivates care with forecast loss when squared returns are noisy risk targets. We score a return second moment with QLIKE.
- usesdoi:10.1080/01621459.1994.10476870
Paired uncertainty estimates use a stationary bootstrap over complete annual blocks.
Cited by
8 publicationsPublic catalog entries on Sciscape.
- usesA Student t(5) model reduced joint forecast loss at equal covariance for three US equity factors in a 2005–2025 annual archive test
We use the retained annual source collection and conservative October activation convention; this report studies a different risk-history question.
- usesA fixed Student t model cut mean joint loss for three US factors in 2005–2025, but the gain was uncertain with single annual blocks
We reuse the official annual archive collection and the conservative October to September forecast schedule.
- usesA six-month risk decay rule gave no clear forecast gain in 20 US market tests in 2005–2025 under assumed archive release dates
We reuse its retained official annual French factor archives and its assumed release schedule. Our risk states, decay rules and primary comparison differ. The 240 market outcomes are shared; this is not independent market evidence.
- usesA one-year weight half-life gave no clear joint forecast gain for three US equity factors in 2005–2025 archive tests
We use its retained official annual archive collection and its July cutoff and October test schedule.
- A t(5) model gave no clear monthly US market density forecast gain with ten years of history in 2005–2025 archive tests
Its risk-history result at the same annual schedule motivates a separate test of distribution shape at fixed variance.
- usesEmpirical US monthly return forecasts gave no clear CRPS gain over normal forecasts in 2005–2025 annual archive tests
We reuse its retained official annual source collection and study a different forecast target on the same market sample.
- usesThree-month blocks gave no clear annual 10% US market loss forecast gain in 20 tests under assumed archive dates, 2005–2025
We reuse the retained annual archive collection and its conservative July input and October start schedule.
- usesCutting estimated correlations in half gave no clear joint forecast gain for three US equity factors in 20 annual archive tests, 2005–2025
We reuse its retained official annual factor archives and its July cutoff and October activation rules. We write new joint forecast and check code.