Quantlake Time in the Market Study: Method and Data

How the Quantlake Time in the Market Study was built: 127 daily series, the benchmark every timing decision is measured against, how a trade is filled, and where the data comes from.

Methodology version 1.0 · data through 31 August 2026 · updated 24 September 2026

If a figure in one of the time-in-the-market papers looks too good or too bad, this page sets out what was tested, how a trade fills, what each decision is compared with, and where the prices come from.

Methodology at a glance

What this page covers

• The instruments, and where their prices come from
• Execution, cash and costs
• The benchmark
• What the study does not claim

The instruments

What was tested. 90 ETFs, 27 mutual funds with long histories and 10 index series, daily, every one of them total return except the Nikkei 225, which has no total-return version from our provider and is labelled wherever it appears. The ETFs and mutual funds are the list used by the Quantlake DCA Study, and the full list, with how it was built, is on Quantlake studies: the ETFs and funds tested, and the method. The index series are these:

Dividends matter to the result. From 2008, the Hang Seng price index returned −0.4% a year. Its total-return index returned 3.15%.

The S&P 500 before 1988 uses the price index with Shiller's monthly dividends spread over each month's sessions. The overlap with the official total-return index ran from 1988 to June 2026. The series returned 11.42% a year against 11.48%.

Execution, cash and costs

How a trade is filled. A signal read at the close of day t is executed at the open of day t+1 for an ETF and at the close of day t+1 for a fund or an index, which is how a fund order fills. Costs are 1 basis point a switch as the base case, with 5 and 25 also reported. Taxes are not modelled.

On an ETF switch day, the overnight gap belongs to the position being left.

Cash. Money out of the market earns the 3-month Treasury bill index less 10 basis points a year, as in the DCA study, and earns nothing before 1954 or for the non-US indices. The published ten-best-days exhibit assumes cash earns nothing. That version is also computed to reproduce it.

The benchmark

The benchmark is buy-and-hold, held from the first day to the last: what an investor who did nothing earned, on the same instrument, over the same dates. Every gap in these papers is that comparison, taken on each instrument against its own buy-and-hold and then read at the median.

A rule that steps out holds the asset for part of the period and Treasury bills for the rest, so it is not comparing like with like on exposure. The share of days each rule held the asset is printed beside its figures rather than adjusted away. The results files also carry a second benchmark, a never-traded portfolio holding the rule's own average share of the asset, which the study's own bar is judged on; the papers do not report it, because its starting split is solved from the share the rule turned out to hold, a number nobody had at the start.

A graded rule is read the same way: it trades only when it moves to a new step, and in between its holding drifts with the market, so its own share of the asset is what gets reported.

Confidence intervals come from a stationary block bootstrap over dates, mean block 252 sessions, drawn jointly across instruments so their co-movement is kept. The random-placement comparison shifts a rule's own in-and-out calendar by one offset shared across instruments, 500 times.

Annual figures are annualised over each market's own number of sessions in a calendar year: about 252 for US series and 245 to 261 for the non-US indices. A rule that reads a trailing window before it can act, such as a day among the ten worst of five or ten years, is scored from the first day that window exists. No measure includes years in which the rule could not trade.

What this study does not claim

• The study reads daily closes, so it cannot see what any individual investor did.
• Funds that closed are absent, so every figure describes funds that survived, under buy-and-hold and under every rule.
• Taxes on dividends and on sales are left out of every figure.

Data

Prices come from EODHD end-of-day data. The S&P 500 before 1988, its dividends, and US inflation come from Robert Shiller's data. Treasury bill rates come from FRED.

Bad prints. A day is flagged when it moves more than 8%, or more than 6 of the instrument's own daily standard deviations, and the next day reverses at least 60% of it. Of the 392 days flagged, 11 were removed on evidence: a print confirmed by hand, or a fund that tracks the instrument closely and did not share the move. 261 were kept because such a fund moved the same way. The other 120 had no fund to check against and were kept as printed. A removed day is dropped rather than repaired, so the return across the two days is kept and no price is invented.

Where it is used

• Missing the 10 Best Days: What That Chart Measures
• Missing the Best Days in Japan, Europe and Since 1929
• Can a Timing Rule Avoid the Worst Days? Three Tested
• What Selling in a Crash and Waiting to Buy Back Cost
• What Followed the Market's Worst Days, and Its Best
• Rolling 10-Year Returns and Ranks Across Our ETFs and Funds

Frequently asked questions

Related

• Missing the 10 Best Days: What That Chart Measures
• Missing the Best Days in Japan, Europe and Since 1929
• Can a Timing Rule Avoid the Worst Days? Three Tested
• What Selling in a Crash and Waiting to Buy Back Cost
• What Followed the Market's Worst Days, and Its Best
• Rolling 10-Year Returns and Ranks Across Our ETFs and Funds
• Quantlake studies: the ETFs and funds tested, and the method