QuanterLab · Research

NASDAQ-100 Additions versus Deletions: Eighteen Decembers, an Effect That Died Around 2016, and the Ticker That Lied

A walk-forward investigation with hypotheses registered before scoring and the size of the search left unreconstructed rather than guessed.

By Serhat Girgin

Universe · Nasdaq 100 (event baskets resolved from the dated index change-log)

Method · Comparative: Additions vs Deletions

Manipulated variable ·

The event side, and nothing else. Both arms read the same dated NASDAQ-100 change record and collect the names whose change is effective within twelve days of the anchor. Arm A holds the additions, arm B the deletions. Each book is bought at the anchor in equal weight, long only, and held one year with no re-selection, because an event basket re-selected at year end would quietly swap into the following year's event. This study anchors each window at the effective day itself. The companion study asks the identical question from the other side of the week: the annual re-ranking is announced on the second Friday of December and takes effect at the open on the Monday after the third Friday, so the two anchor schedules differ by five trading days, and the difference between the studies is the price of that week. Anchor dates are rule-derived and verified against the exchange's December press releases rather than taken from the data vendor, whose date field records the announcement week in some years and the effective week in others. Both arms are measured in total returns with dividends credited at the ex-date, ten basis points charged per one-way traded dollar, against the NASDAQ-100 tracker QQQ rebuilt as a total-return index. The baskets are a pure function of the anchor and the public change record: nothing here is screened, ranked or chosen by the authors. Additions occasionally include new share classes of companies already in the index; they are kept, because excluding them would be a choice, and disclosed. Deletions that the data vendor no longer serves (delisted or acquired names) cannot be held; every such refusal is counted in the record. Refusals also cover a recycled symbol whose vendor series demonstrably belongs to a later company (the identity audit is published with the paper). A window enters the study only if at least two names on each side carry price data at its anchor; 2006 fails that floor, with one priceable deletion, and is excluded, leaving eighteen windows from 2007 to 2024. A held name whose series ends inside the window, an acquisition closing mid-year, is carried at its final print for the remainder of the window.

Step size · 1 year per forward window

Out-of-sample span · 2007-12-24 → 2025-12-23

Compiled · August 26, 2026

Search record · none (size unknown, see §2.3)

Abstract

Every December from 2007 to 2024, the NASDAQ-100 announced its annual re-ranking on the second Friday and applied it at the open on the Monday after the third. This study bought both sides of that event, every year, and held each side for one year: the additions and the deletions, each at two moments, the day after the announcement and the effective day itself. Every basket is a pure function of the index's dated change record. Nothing here is screened or chosen; the rulebook picks every portfolio, and every window was registered before it ran.

Pooled over eighteen years the deletions beat the additions by 9.2 points a window, and in the spirit of Arnott's index-fund work that is the expected sign. But the pooled number is not the finding. Split the record at 2016 and the effect falls apart in the open: before 2016 the excluded stocks beat the included by 26.6 points a window and won eight years of nine; from 2016 on the additions were ahead by 8.2 and won six of nine. The deletion premium, worth 18.7 points a window over the index itself in the first era, ran 16.3 points behind it in the second. The recovery trade the December machine used to hand out, Sirius at four times, Netflix at four times, stopped paying around the same time the index-effect literature says the announcement pop died.

One thing did not move between the eras: measured against capitalization-weighted QQQ, the additions lagged by about eight points a window in both halves, though against an equal-weight ruler most of that lag is weighting, not the event.

And one result in this paper exists because an audit killed a better one. The data vendor serves Overstock's price history under Bed Bath and Beyond's recycled ticker, and that impostor, passed by every shape-based gate, was briefly enough to make the deletions beat the index outright. The version you are reading survived its own audit. The more exciting version did not.

The single-channel December regime this record measures ended in May 2026, when quarterly membership reviews and a fast-entry lane joined it. Whatever replaces it, 2007 to 2024 is the clean record of the rule as it was.

1 Methodology

WHERE THIS SITS

Two literatures ask what an index change does to a stock, and they ask different questions. The first measures the days around the event: Shleifer (1986) and Harris and Gurel (1986) used S&P 500 additions to ask whether demand curves for stocks slope down, Chen, Noronha and Singal (2004) documented that additions and deletions do not mirror each other, and Greenwood and Sammon (2025) showed the addition-day pop has decayed from over seven percent in the 1990s to roughly nothing. Biktimirov and Xu (2019) brought that machinery to the NASDAQ-100 and found days-scale effects there too.

The second literature asks what happens afterwards. Arnott, Kalesnik and Wu (2018) showed that capitalization-weighted indices systematically add stocks after they have outperformed and delete them after they have underperformed, so an index fund is structurally a buyer of the expensive and a seller of the cheap, and the reversal that follows is long and slow.

This study belongs to the second literature and applies it where it has not been applied. The long-horizon work is built on the S&P 500 and broad cap-weighted families; the NASDAQ-100 work measures days. Nobody had asked the one-year question of the NASDAQ-100, an index with a date-based rulebook rather than committee discretion, a technology-heavy base, and one fixed December reconstitution. And the question has a deadline attached: since May 2026 the December event is no longer the sole regular entry channel, with membership now also moving at quarterly reviews and through a fast-entry lane (Nasdaq, 2026). The windows here, 2007 through 2024, are the complete record of the single-channel December regime.

HOW THE DECEMBER MACHINE WORKED

The NASDAQ-100 holds roughly one hundred of the largest non-financial companies listed on Nasdaq, weighted by a modified capitalization scheme. Once a year the index re-ranked its universe, announced the resulting additions and deletions on or about the second Friday of December, and applied them at the open on the Monday after the third Friday. The week in between is the most telegraphed trade in markets: everyone knows what the trackers must buy and sell, and when.

That calendar is the study's clock. Anchor dates are derived from the rule itself and checked against the exchange's December press releases, not taken from the data vendor, whose date field records the announcement week in some years and the effective week in others. The addition book and the deletion book are collected mechanically: every name whose change is effective within twelve days of the anchor, from the same dated change record the platform's point-in-time membership machinery already uses. Occasional off-cycle December events collected by that rule stay in, disclosed; the largest is Facebook's special addition in December 2012.

FOUR BOOKS, ONE RULER

Each December, four portfolios form: additions and deletions, each bought at the announcement timing (first trading day after the announcement) and at the effective timing (the Monday the change takes force). Equal weight, long only, held one year untouched, because an event basket re-selected at year end would quietly roll into the next year's event.

Every return is a total return, dividends credited on the ex-date and held as cash. Ten basis points come off every dollar traded, in all books alike. The benchmark is QQQ, the index's own tracker, rebuilt as a total-return series from its distribution record; because the books are equal-weight and QQQ is capitalization-weighted, the findings also read the books against the equal-weight tracker QQQE for the windows it has existed. The platform's comparison machinery is pairwise, so the record is sealed as two registered studies, additions against deletions at each timing, sharing one event table; this page hosts the effective-day pair and carries the announcement pair through every table as companion books.

The sealed, pre-registered contrast is additions against deletions. Statements about either book against the index are descriptions of realized paths, not registered claims, and they carry the weighting caveat above.

A name whose series ends inside a window, an acquisition closing mid-year, is carried at its final print for the remainder. A window enters the study only if at least two names on each side carry price data; 2006 fails that floor with one priceable deletion, leaving eighteen windows, 2007 through 2024.

2 Results

2.1 Headline

Additions, pooled Sharpe

0.42

4531 OOS bars

Deletions, pooled Sharpe

0.67

4531 OOS bars

The statistic this paper stands on

Additions led 7 of 18 windows; a sign test reads p = 0.24. Before 2016 deletions won by 26.6 points a window and took eight years of nine; from 2016 additions led by 8.2. The era split, not the pooled gap, is the finding.

show or hide a line: Additions Deletions platform reference (SPY)

Figure 1. Both arms stitched through the identical windows, Additions (+313.6%), Deletions (+1536.9%), platform reference grey (+1591.5%, total return, its own dividends reinvested, pooled Sharpe 0.814). Dotted verticals mark the step boundaries; the dashed horizontal is break-even. These figures compound each arm's own stitched daily series; the pooled statistics in the text inner-join both arms' trading days, one session apart, both are printed from the frozen record.

**Table 1.** Every book in this study on one ruler
Book	Total return	Per year	Volatility	Sharpe	Worst drawdown	Mean window
Deletions, announcement timing (companion run)	+1750.4%	17.60%	30.0%	0.69	-56.6%	+22.58%
Benchmark (reference)	+1591.5%	17.01%	22.4%	0.81	-51.2%	+20.42%
Deletions	+1536.9%	16.80%	30.0%	0.67	-56.6%	+21.73%
Additions	+313.6%	8.21%	29.7%	0.41	-53.2%	+12.52%
Additions, announcement timing (companion run)	+301.5%	8.03%	29.4%	0.41	-53.2%	+11.39%

Volatility, Sharpe and worst drawdown are computed on each book's own stitched daily series over the identical trading days that Figure 1 draws, so the panel and the figure are the same arithmetic. Sharpe carries no cash hurdle. Mean window is the arithmetic average of the one-year window returns and does not compound to the total beside it; the difference is volatility drag.

Figure 2. The measurement ladder: Additions's whole walk, chained five ways. price only (+279.8%), with dividends (+321.2%), net of costs (+313.6%), against the benchmark measured both ways: price only (+1373.5%), total return (+1591.5%). The distance between the two green pairs is the dividends collected; the sliver between the last two greens is the cost bill; the distance between the two greys is what a price-only chart hides about the index. Every other figure on this page uses the deepest rung on each side, net of costs against the total-return benchmark.

Figure 3. The same walk, re-based to 1.00x at the first window starting in 2016, 9 of the 18 windows above. Additions (+223.1%), Deletions (+116.8%), benchmark grey (+536.1%, total return). This is a subset of Figure 1, not a correction to it. The era boundary here is pinned by the author at 2016 rather than left at the platform default, and the era rows below put a number on the two periods it separates. Whether the record actually breaks there is a question those rows answer, not one this caption settles.

**Figure 4.** The headline under a leave-one-window-out test. The pooled 9.2 point gap rests heavily on the 2008 window; the median of the eighteen gaps is 5.8 and the trimmed mean the same.

**Figure 5.** Each book against its index, era by era, and against the equal-weight ruler where one exists. The vs-index readings are descriptive; the registered contrast is additions against deletions.

2.2 Per-step results

**Table 2.** One row per step, raw out-of-sample results. A short window can pair a negative return with a positive annualised Sharpe: at high daily volatility the arithmetic mean of daily returns sits above the compounded window return, and the Sharpe reads the former. Volatility drag, printed rather than smoothed.
#	Out-of-sample window	Additions SR	Deletions SR	Additions, announcement timing SR	Deletions, announcement timing SR
1	2007-12-24 → 2008-12-23	-0.95	-0.23	-0.71	-0.05
2	2008-12-22 → 2009-12-22	1.44	2.01	1.56	2.09
3	2009-12-21 → 2010-12-21	0.78	1.85	0.75	1.86
4	2010-12-20 → 2011-12-20	-0.78	0.09	-0.91	0.08
5	2011-12-19 → 2012-12-18	0.47	0.89	0.39	0.36
6	2012-12-24 → 2013-12-24	2.50	2.77	2.38	2.46
7	2013-12-23 → 2014-12-23	1.26	0.15	1.25	0.12
8	2014-12-22 → 2015-12-22	-0.27	0.64	0.06	0.86
9	2015-12-21 → 2016-12-20	-0.56	0.40	-0.41	0.73
10	2016-12-19 → 2017-12-19	1.74	1.44	1.50	1.61
11	2017-12-18 → 2018-12-18	0.48	0.43	0.67	0.94
12	2018-12-24 → 2019-12-24	2.36	0.89	1.85	0.57
13	2019-12-23 → 2020-12-22	0.99	0.24	0.93	0.28
14	2020-12-21 → 2021-12-21	0.47	0.71	0.60	0.57
15	2021-12-20 → 2022-12-20	-0.98	0.43	-0.88	0.62
16	2022-12-19 → 2023-12-19	1.06	0.99	0.62	0.76
17	2023-12-18 → 2024-12-17	0.63	0.22	0.88	0.50
18	2024-12-23 → 2025-12-23	0.72	0.14	0.56	-0.00

Figure 6. Additions: every step's out-of-sample curve overlaid, each rebased to 1× at its own start. Read alongside Table 1: consistent shape across steps is the walk-forward's evidence; a single lucky leg is not.

Figure 7. Deletions: the same windows, the other arm. Compare shape-for-shape with the previous figure: the two arms trade the identical out-of-sample legs.

2.3 Search accounting

No search record exists for this design. It was not promoted from a recorded evolving search, so the number of alternatives tried before it, on paper, in another tool, or in the author's head, is unknown. Unknown is a different fact from one: a study with no lineage is not a strategy with one trial, it is a strategy with an unrecorded number of them. Accordingly no count of alternatives tried is claimed, and nothing in this paper is corrected for a search that was never recorded; the number that stands is the raw out-of-sample result plus this disclosure. The registered per-step record below (§4) still guarantees each window's hypothesis was hashed and registered before that window was scored.

2.4 The comparison

Both arms trade the same registered windows, so their returns can be PAIRED: inside each window the two return series are inner-joined date by date and the difference r_Additions − r_Deletions is the object under test. Because this is ONE pre-declared contrast, frozen at registration before any window was scored, the paired statistic needs no multiple-testing deflation, and the per-arm pooled numbers above are likewise uncorrected, this design has no recorded search to correct against (§2.3). The paired contrast carries that gap differently from the arm levels: it was declared before each window ran and it is scored on the difference, so no window could be read first and scored after. What the timestamps cannot speak to is how this contrast came to be the declared one, which is the limit §2.3 states.

**Table 3.** Window-by-window paired comparison. Δ is the growth gap (Additions − Deletions) over the window's paired dates. The Additions, announcement timing and Deletions, announcement timing columns carry companion books' growth over the full window; the Δ and Leader columns compare only this study's two registered arms.
#	Window	Paired bars	Additions	Deletions	Δ	Leader	Additions, announcement timing	Deletions, announcement timing
1	2007-12-26 → 2008-12-23	252	-43.7%	-21.6%	-22.1 pp	Deletions	-36.8%	-14.3%
2	2008-12-23 → 2009-12-22	252	+46.2%	+170.1%	-123.9 pp	Deletions	+52.7%	+183.4%
3	2009-12-22 → 2010-12-21	252	+13.8%	+47.3%	-33.5 pp	Deletions	+13.1%	+47.8%
4	2010-12-21 → 2011-12-20	253	-21.5%	-1.5%	-19.9 pp	Deletions	-24.2%	-1.7%
5	2011-12-20 → 2012-12-18	250	+11.0%	+20.2%	-9.3 pp	Deletions	+7.1%	+5.9%
6	2012-12-26 → 2013-12-24	252	+46.8%	+71.8%	-25.0 pp	Deletions	+43.7%	+62.6%
7	2013-12-24 → 2014-12-23	252	+32.3%	+1.1%	+31.2 pp	Additions	+31.3%	+0.6%
8	2014-12-23 → 2015-12-22	252	-6.4%	+12.3%	-18.7 pp	Deletions	-0.7%	+17.9%
9	2015-12-22 → 2016-12-20	252	-12.9%	+5.7%	-18.6 pp	Deletions	-10.3%	+12.1%
10	2016-12-20 → 2017-12-19	252	+25.3%	+25.9%	-0.6 pp	Deletions	+21.1%	+29.6%
11	2017-12-19 → 2018-12-18	251	+9.4%	+7.2%	+2.2 pp	Additions	+14.3%	+19.8%
12	2018-12-26 → 2019-12-24	252	+64.0%	+13.4%	+50.6 pp	Additions	+47.8%	+6.7%
13	2019-12-24 → 2020-12-22	252	+35.3%	+2.2%	+33.1 pp	Additions	+32.1%	+3.6%
14	2020-12-22 → 2021-12-21	252	+9.1%	+11.4%	-2.3 pp	Deletions	+12.3%	+8.3%
15	2021-12-21 → 2022-12-20	252	-42.3%	+7.2%	-49.5 pp	Deletions	-39.6%	+11.5%
16	2022-12-20 → 2023-12-19	251	+23.5%	+22.9%	+0.7 pp	Additions	+11.9%	+16.6%
17	2023-12-19 → 2024-12-17	251	+8.5%	+2.4%	+6.1 pp	Additions	+12.3%	+9.7%
18	2024-12-24 → 2025-12-23	250	+27.1%	-6.9%	+34.0 pp	Additions	+17.0%	-13.6%

Paired Sharpe of the difference track: -0.28 · block bootstrap (2000 paths, block 10, seed 1234): P(Additions beats Deletions) = 11.8%.

Window win-rate. Additions led 7 of 18 windows (38.9%), Deletions led 11, and the mean window gap of -9.20 pp points the same way. Widest single window: 2008 at -123.9 pp.

**Table 4.** The same comparison split at 2016. Pooling the whole walk into one row hides which side of the split the difference came from.
Period	Windows	Additions	Deletions	Mean gap	Additions led
All windows	18	+12.53%	+21.73%	-9.20 pp	7/18
Before 2016	9	+7.29%	+33.93%	-26.64 pp	1/9
2016 onward	9	+17.77%	+9.52%	+8.24 pp	6/9

Figure 8. Mean window return per period. Additions above, Deletions below, with the gap at right. The pooled bar and the post-2016 bar are the same comparison over different periods.

The two eras disagree by 34.88 pp. The pooled figure is therefore not a standing property of either method, it is dominated by the earlier period. Read the two rows, not the average.

3 The circuit

The strategy is a circuit of platform primitives, frozen when the study is registered. Below is the circuit as wired on the canvas, the objective it encodes and how the search runs through it, followed by the mathematics each primitive actually computes, the same formulas the execution engine runs. The complete parameterisation is preserved in the study ledger (Appendix A).

The hypothesis under test

A COMPARATIVE study: Additions vs Deletions, walked on the same registered out-of-sample windows. Additions: A configured universe, held as a buy-and-hold basket, and run out-of-sample from the anchor: whatever the design estimates from history is re-estimated at each anchor from pre-anchor data only, and the walk advances through registered out-of-sample windows; its disposition is the realized forward path versus the benchmark. Deletions: A configured universe, held as a buy-and-hold basket, and run out-of-sample from the anchor: whatever the design estimates from history is re-estimated at each anchor from pre-anchor data only, and the walk advances through registered out-of-sample windows; its disposition is the realized forward path versus the benchmark. The arms differ in: Index Event Basket, event: additions → deletions. The contrast under test: whether Additions generates better risk-adjusted returns than Deletions over the identical out-of-sample windows.

Figure 9. The frozen circuit, every node a primitive, every wire a typed data-flow; the two arms are colour-coded (Additions green, Deletions blue, shared feeds neutral). Each box is one step of the strategy; data flows along the wires left to right, and no box can see data dated later than the box feeding it. The whole diagram was frozen when the hypothesis was registered. Click any node to open what that step ran with and what it produced.

Envelopes show counts, ratios, dates, and the parameters the author chose. Full price and per-name data series are not republished: the underlying market data is licensed to QuanterLab. Point figures quoted in the prose, a named holding's return over a stated span, are summary facts derived from public market prices, not redistributed series.

What each part does

Portfolio Backtest, Replay the portfolio forward, rebalanced, point-in-time, with costs.

Transaction Cost, Charge for trading, slippage + commission on every turn.

Portfolio Forward Autopsy, The post-mortem, where the forward test’s return actually came from.

The objective and the search

Additions

Universe	No explicit universe defined.
Validation & out-of-sample	portfolio forward test (buy-and-hold book) (1y horizon from the anchor, held (no rebalance)); overlays: Transaction Cost.

Deletions

The specification is identical to Additions's table above, row for row; the one sealed difference between the arms is itemized below.

What differs between the arms, one difference; the comparison is clean:

paramIndex Event Basket, event: additions → deletions

Companion book. Additions, announcement timing comes from a companion run of the identical configuration on its own sealed anchor schedule. Its results run through this paper's tables under its own name.

Companion book. Deletions, announcement timing comes from a companion run whose one change against this study's Additions arm is Index Event Basket, event: additions → deletions. Its results run through this paper's tables under its own name.

Everything else is held identical, so an out-of-sample gap between the arms is attributable to this one change.

Cost elements are wired into the circuit, the realised drag is reported per step in Appendix B.

Show the mathematics, 3 primitives, formulas and parity notes

3.1 Portfolio Backtest

Replay the portfolio forward, rebalanced, point-in-time, with costs.

Holds the basket and rebalances on schedule, re-selecting and re-optimizing point-in-time at each rebalance (so it only ever uses information available then), and reports the equity curve, Sharpe, drawdown and trade stats, optionally net of cost and risk overlays.

Compounded equity

E_t = E_{t-1}\big(1 + \mathbf w_{t}^{\top}\mathbf r_t - \text{costs}_t\big)

Drawdown

\text{DD}_t = \frac{E_t}{\max_{\tau\le t}E_\tau} - 1, \qquad \text{MaxDD} = \min_t \text{DD}_t

Financing a levered book

\text{charge}_t \;=\; \text{loan}_t \cdot \frac{\text{spread}}{252}, \qquad \text{loan}_t = \begin{cases}\max(0,\,-\text{cash}_t) & \text{institutional (netted)}\\ \max(0,\,\text{long MV}_t - E_t) & \text{retail (no netting)}\end{cases}

A levered long/short book (the β-neutral Long/Short Select) borrows its excess notional. WHO you are decides the loan: a prime broker nets short-sale proceeds against the margin loan, a fully-netted BAB book carries almost none, while a retail margin account cannot net, so the same book borrows the long leg’s excess over equity. Profiles: institutional = 50 bps spread + 25 bps GC short borrow; retail = 350 bps + 150 bps (the BEST retail tier, so any verdict is conservative); custom = your own knobs, spread on negative cash. Charged daily, reported as financing_drag_pct, never silent.

3.2 Transaction Cost

Charge for trading, slippage + commission on every turn.

Real trading isn't free. This deducts a cost proportional to how much you trade (turnover), in basis points, so the backtest reflects net, not gross, performance.

Cost per rebalance

\text{cost}_t = \frac{\text{bps}}{10{,}000}\;\times\;\text{turnover}_t, \qquad \text{turnover}_t = \tfrac12\sum_i \lvert w_{i,t}-w_{i,t^-}\rvert

3.3 Portfolio Forward Autopsy

The post-mortem, where the forward test’s return actually came from.

Runs after the Portfolio Forward Test and dissects its realized path: per-rebalance contributions, winners and losers, exposure and cash periods, and how the realized route compares to what the risk cones projected. It computes nothing new about the future, it explains the past the book just lived.

Reading it

Depth I–IV: headline attribution, per-segment breakdown, per-name contributions, and the calibration ledger (projected cone vs realized, segment by segment). In a study, this is the node that fills the appendices.

4 Projection calibration, pooled across the walk

Every rebalance carried a Monte Carlo cone and a 95% VaR estimated before the segment it is scored against. Two questions, pooled over the whole study: did realized outcomes land inside the band as often as the band claims, and were VaR breaches as frequent as 5%?

This section is produced by the forward tester itself: every portfolio backtest fits the cone and the VaR estimate at each rebalance and scores them against the segment that followed. It does not require, and this circuit does not contain, a Monte Carlo primitive; that primitive is a separate, standalone analysis.

Figure 10. Projected range versus what occurred, at each of 72 scored rebalance segments, pooled across both arms. The final rebalance of each step has no following segment to score, the ledger marks those rows “no segment follows this rebalance”, which is why this count sits below the raw rebalance totals in the table beneath. Each vertical bar is that rebalance's P5–P95 Monte Carlo cone with the median ticked; the dot is the realized return of the segment that followed. Filled green = the outcome landed inside its own cone; red = it did not. The strip beneath repeats that as one mark per rebalance, so a run of misses in one period is visible as a run. Every cone was fitted only on data prior to the segment it is scored against.

Arm	Steps	Rebalances	In band	Coverage	Expected	VaR days	Breach rate	Expected
Additions	18	36	29 / 36	80.6% ±6.6	90.0%	4418	6.77% ±0.378	5.0%
Deletions	18	36	32 / 36	88.9% ±5.24	90.0%	4512	4.3% ±0.302	5.0%
Additions, announcement timing (companion run)	18	36	29 / 36	80.6% ±6.6	90.0%	4413	6.78% ±0.378	5.0%
Deletions, announcement timing (companion run)	18	36	32 / 36	88.9% ±5.24	90.0%	4512	4.23% ±0.3	5.0%

Note. The projection bands here are estimated from each basket's own trailing two years, and these baskets hold three to seven names, so the bands are extremely wide and inherit whatever those two years did; a cone whose fifth percentile sits above zero after a recovery year is the estimator, not the market. Single-window VaR breach rates near twenty percent are the same small-book arithmetic. The displacement runs one way, so the coverage figures above are a floor, and no claim in the paper rests on this section.

± values are binomial standard errors on the estimate. A coverage figure below the expected band means the projection was over-confident; a breach rate above 5% means the same of the risk model. Both forecasts used only data prior to the segment scored.

5 Discussion

5.1 Findings

THE EFFECT, AND WHEN IT DIED

Pooled over all eighteen windows, the deletions beat the additions by 9.20 points a window. That number is true and it is not the finding, because it does not survive its own record. The median of the eighteen gaps is 5.8 points; drop the single 2008 window, where the deletions' post-crisis recovery beat the additions by 124 points, and the pooled gap falls to 2.5. The figure below shows every version of the headline with one window removed.

The era split is the summary that survives, and it does not depend on where the knife falls. Before 2016 the deletions won eight windows of nine, by 26.6 points a window on average. From 2016 on the additions won six of nine, by 8.2. Move the split anywhere in the middle of the decade and the same shape appears: cut at 2014 and the halves read minus 28.9 then plus 3.3; at 2015, minus 27.6 then plus 5.5; at 2017, minus 24.0 then plus 9.3; at 2018, minus 21.6 then plus 10.4, with additions leading most windows after every candidate boundary. The paper does not need 2016; it needs the mid-2010s, and any cut through them says the same thing. Windows are named by their December anchor throughout, so the 2008 window is calendar 2009, and so on. The second era's lead is itself fragile, resting largely on 2018's AMD-led class; drop that window and it shrinks to three points. The defensible statement is not that the effect reversed. It is that a large early effect faded to roughly a coin flip, on the same clock on which Greenwood and Sammon (2025) show the announcement-day effect decaying to nothing. The splash died, and so did the tide.

Window counts say the same thing without any pooling. The additions led 7 of 18 windows; a sign test puts the one-sided probability of that under a fair coin at 0.24, which is a lean and nothing stronger. The platform's daily block bootstrap across 4,530 paired bars reads 11.8 percent at this timing and 8.8 at the announcement timing, but eighteen annual events are eighteen observations, and the window-level test is the one this paper stands on.

WHAT DIED, PRECISELY

Split each book against the index and the decay localizes. The deletions beat QQQ by 18.7 points a window before 2016, winning five windows of nine, carried by the recovery trades the December machine used to sell into: the 2008 class returned 170 percent at this timing with Sirius XM up roughly 390 percent, and the 2012 class 72 percent with Netflix up roughly 310. The pooled Sharpe of the deletion book, 0.67 at this timing and 0.69 at the announcement, sits below QQQ's 0.81 in both eras taken together. From 2016 on the deletions ran 16.3 points a window behind the index and beat it twice in nine years. The resurrection windfalls simply stopped arriving.

The additions' shortfall against QQQ, by contrast, barely moved: 7.9 points a window before 2016, 8.1 after. But that constancy is partly an artifact of the ruler. Against the equal-weight tracker QQQE, which exists from 2012, the additions' lag largely disappears (16.8 a window against QQQE's 17.4 over those thirteen windows) while the deletions run 3.9 behind. The era table below the leave-one-out figure puts every one of these vs-index readings in one place. Much of the books' shortfall against QQQ in the second era is mega-cap concentration, not the event. The registered contrast between the two books is immune to that caveat; the vs-index readings are not, and both are shown.

THE WEEK EVERYONE TRADES

Buying deletions at the announcement rather than the effective day added 0.98 points to the average window, positive in 10 of 18 years. Buying additions early subtracted 1.01, positive in 8. One point a year, signs split near half, on books whose windows swing by forty: whatever the index-arbitrage week pays the people who trade it in hours, a one-year holder cannot collect it.

THE TICKER THAT LIED

This paper's strongest result is one it does not contain. The first sealed walk of this study reported that the deletions beat the index outright. The per-name audit that followed found the reason: the vendor's price series for BBBY, a 2016 deletion, is not Bed Bath and Beyond. Overstock dot com bought the brand out of the 2023 bankruptcy and took the ticker and the name, and the vendor serves the buyer's history under the dead company's symbol, profile included. Held unknowingly at one quarter of the four-name 2016 book, the impostor printed plus 269 percent for a year in which the real Bed Bath fell by a third.

Every shape-based data gate passed it, because it is a real company's real prices. It was caught by checking identities, name by name, against the change record, and the platform now refuses that class of series by an evidence-backed denylist. The corrected 2016 deletion window is plus 26 percent instead of plus 94, and the corrected deletion book straddles the index instead of beating it. The finding you are reading survived its own audit; the more exciting version of it did not.

WHAT COULD NOT BE HELD

Fifteen addition slots and fourteen deletion slots, of 243, could not be held: names the vendor no longer serves because they were acquired, delisted, or renamed out of their ticker. Each refusal is named in the appendix record of its window. The direction of the resulting bias is genuinely unknowable from stored data, because the missing deletions include both acquisition exits, which usually left at a premium, and collapses like Sears, which did not.

5.2 Interpretation

A NASDAQ-100 deletion was not an opinion about a business. It was an arithmetic consequence of a ranking: the index holds the hundred largest, so the stocks that left were, almost by construction, the ones whose prices had already fallen furthest, and the stocks that arrived were the ones whose prices had already risen most. The December machine was a rule for selling last year's losers to buy last year's winners, executed once a year, in public, on a printed calendar.

Arnott, Kalesnik and Wu (2018) argued that this is what capitalization-weighted indexing quietly does everywhere. The NASDAQ-100 through 2024 was the cleanest possible place to watch it, because there was no committee judgment to argue about, only arithmetic and a calendar. And watched over eighteen years, the record says two things at once.

First, the reversal Arnott's line of work predicts was there, and it was enormous, while it lasted. Through 2015, buying the December castoffs beat buying the December darlings by 26.6 points a window, and beat the index itself by 18.7. Second, it faded. From 2016 the castoffs stopped recovering, the darlings stopped underperforming them, and the contrast settled into noise with a mild lean the other way. The literature's announcement-day effect and this record's one-year effect decayed on the same clock, which is at least consistent with one cause: a trade that became famous, crowded, and then gone.

What survives every cut is duller and still worth money: at either timing, in either era, the newly included stocks were not the free ride their headlines suggested, lagging the capitalization-weighted index they had just joined by about eight points a window, though against an equal-weight ruler most of that is weighting rather than the event. An investor who greeted each December class with a year of patience gave up nothing measurable at any point in the record.

In May 2026 the December machine lost its monopoly: membership now also moves at quarterly reviews, and a fast-entry lane admits qualifying companies between them. Whether spreading the same arithmetic across smaller, more frequent events revives the premium or buries it deeper is a question someone can ask in fifteen years. The single-channel record is closed, and this is what it says.

No search record exists for this study: the design was not promoted from a recorded evolving search, so the number of alternatives tried before it is UNKNOWN, which is a different fact from one. No count of alternatives tried is claimed, and nothing is corrected for a search that was never recorded; the number that stands is the raw out-of-sample result plus this disclosure. The out-of-sample windows are historical.

5.3 Limitations

Small books are loud. The yearly classes run three to eleven names, so single stocks decide single windows, and both eras' summaries are fragile to their largest years: the pre-2016 gap leans on 2008, the post-2016 lean on 2018. The leave-one-out figure and the era table exist so that dependence is visible rather than smoothed away, and the window-level sign test, not the daily bootstrap, is the statistic this paper stands on.

The registered contrast is additions against deletions. Every statement against an index is descriptive, and the choice of ruler matters: the books are equal-weight, QQQ is capitalization-weighted, and the equal-weight tracker QQQE only exists from 2012, so the first five windows have no like-for-like index reading at all.

Twenty-nine of 243 slots could not be held, and the direction of that bias is not determinable: the unpriceable deletions include premium acquisition exits and outright collapses. Ticker recycling is a further vendor hazard the audit found once, BBBY, and refused; renamed tickers such as Monster's cost two early addition slots for the opposite reason, the old symbol no longer resolving at the old date. The two timings collect events by the same twelve-day rule around different anchors, so their books can differ at the edges; one 2011 addition enters the announcement study only. Share-class additions, Comcast's and Liberty's K shares among them, are kept, because excluding them would be a choice, and are named in the record.

The projection bands in the calibration section are estimated from each book's own trailing two years, and on three-to-seven-name books they are grotesquely wide and displaced by whatever those two years did; single-window VaR breach rates near twenty percent are the same small-book arithmetic. No claim in this paper rests on them, and the note under the calibration table says the same where the reader meets it.

Costs are charged on the books and not on the index, the standard convention. Anchor dates are rule-derived and checked against press releases; the vendor's own date field could not be used for timing. Eighteen windows are eighteen observations of an annual event, the regime that produced them no longer runs alone, and the effect this record measures was already dying inside it.

References

QuanterLab reference architecture

Gelman, A., & Loken, E. (2013). The garden of forking paths: Why multiple comparisons can be a problem, even when there is no “fishing expedition.” Working paper, Columbia University.
Harvey, C. R., Liu, Y., & Zhu, H. (2016). … and the Cross-Section of Expected Returns. Review of Financial Studies, 29(1), 5–68. doi:10.1093/rfs/hhv059
Lo, A. W. (2002). The Statistics of Sharpe Ratios. Financial Analysts Journal, 58(4), 36–52. doi:10.2469/faj.v58.n4.2453

Author’s references?

Arnott, R., Kalesnik, V. and Wu, L. (2018). Buy High and Sell Low with Index Funds! Research Affiliates.
Shleifer, A. (1986). Do Demand Curves for Stocks Slope Down? Journal of Finance 41(3), 579-590.
Harris, L. and Gurel, E. (1986). Price and Volume Effects Associated with Changes in the S&P 500 List. Journal of Finance 41(4), 815-829.
Chen, H., Noronha, G. and Singal, V. (2004). The Price Response to S&P 500 Index Additions and Deletions: Evidence of Asymmetry and a New Explanation. Journal of Finance 59(4), 1901-1929.
Greenwood, R. and Sammon, M. (2025). The Disappearing Index Effect. Journal of Finance 80(2), 657-698.
Biktimirov, E. N. and Xu, Y. (2019). Asymmetric stock price and investor awareness reactions to changes in the Nasdaq 100 index. Journal of Asset Management 20(2), 134-145.
Nasdaq (2026). Nasdaq-100 Index Methodology and the February 2026 consultation on reconstitution changes, effective May 2026. Nasdaq Global Indexes.

Appendix A Reproducibility in QuanterLab

Each step is backed by a frozen run report. The study is re-derivable from the ledger below.

#	Commit	Report	Anchor	OOS window
1	476188644296	1632	2007-12-24	2007-12-24 → 2008-12-23
2	cce5fff72bbb	1633	2008-12-22	2008-12-22 → 2009-12-22
3	8eb4de4d5234	1634	2009-12-21	2009-12-21 → 2010-12-21
4	72cdc54e34fe	1635	2010-12-20	2010-12-20 → 2011-12-20
5	2aa31760def5	1636	2011-12-19	2011-12-19 → 2012-12-18
6	841ef77be818	1637	2012-12-24	2012-12-24 → 2013-12-24
7	02fe6c9580f9	1638	2013-12-23	2013-12-23 → 2014-12-23
8	05041e56de5f	1639	2014-12-22	2014-12-22 → 2015-12-22
9	0136ab8062a6	1640	2015-12-21	2015-12-21 → 2016-12-20
10	19285ed51ad5	1641	2016-12-19	2016-12-19 → 2017-12-19
11	2dccb9ba5316	1642	2017-12-18	2017-12-18 → 2018-12-18
12	efc3488fc7bb	1643	2018-12-24	2018-12-24 → 2019-12-24
13	d6ea4643529b	1644	2019-12-23	2019-12-23 → 2020-12-22
14	9f3ed920b55d	1645	2020-12-21	2020-12-21 → 2021-12-21
15	2a98baf58d38	1646	2021-12-20	2021-12-20 → 2022-12-20
16	23b2425b1479	1647	2022-12-19	2022-12-19 → 2023-12-19
17	24cfe444144a	1648	2023-12-18	2023-12-18 → 2024-12-17
18	b5100fdb0055	1649	2024-12-23	2024-12-23 → 2025-12-23

Appendix A2 Registration record

What this record does and does not establish. Every window in this study is historical: the data existed before the study began, so this is sequential sealing on past windows, not pre-registration in the clinical-trial sense, and no procedure could make it so. What the platform does enforce is order, each step's specification was frozen and hashed before that step was scored, and the walk cannot advance past a step that was never run or close one with a result registered for a different window. The two timestamp columns below are the evidence: read them together and each seal precedes its own run, and each run precedes the next seal. A study whose seals all post-date its runs would show it here. Wall-clock spacing between seals varies with the author's schedule and queue latency; the ordering, not the tempo, is the claim.

“A COMPARATIVE study: Additions vs Deletions, walked on the same registered out-of-sample windows. Additions: A configured universe, held as a buy-and-hold basket, and run out-of-sample from the anchor: whatever the design estimates from history is re-estimated at each anchor from pre-anchor data only, and the walk advances through registered out-of-sample windows; its disposition is the realized forward path versus the benchmark. Deletions: A configured universe, held as a buy-and-hold basket, and run out-of-sample from the anchor: whatever the design estimates from history is re-estimated at each anchor from pre-anchor data only, and the walk advances through registered out-of-sample windows; its disposition is the realized forward path versus the benchmark. The arms differ in: Index Event Basket, event: additions → deletions. The contrast under test: whether Additions generates better risk-adjusted returns than Deletions over the identical out-of-sample windows.”

The same hypothesis was registered independently at every step, hashed before each step's out-of-sample window was scored:

**Table 5.** Registration audit, one row per registered step, with the time each specification was frozen and the time its window was scored. The hypothesis is identical on every row by design: it was registered once and re-registered unchanged at each anchor. Rows that differ would mean the specification moved mid-walk, which is the thing this record exists to rule out. The timestamps are the separate claim: each seal precedes its own run, and each run precedes the next seal.
#	Anchor	Registered at (UTC)	Run completed (UTC)
1	2007-12-24	2026-08-26 17:30:41	2026-08-26 17:31:14
2	2008-12-22	2026-08-26 17:31:16	2026-08-26 17:31:36
3	2009-12-21	2026-08-26 17:31:38	2026-08-26 17:31:48
4	2010-12-20	2026-08-26 17:31:50	2026-08-26 17:32:00
5	2011-12-19	2026-08-26 17:32:03	2026-08-26 17:32:13
6	2012-12-24	2026-08-26 17:32:15	2026-08-26 17:32:35
7	2013-12-23	2026-08-26 17:32:37	2026-08-26 17:32:47
8	2014-12-22	2026-08-26 17:32:49	2026-08-26 17:32:59
9	2015-12-21	2026-08-26 17:33:01	2026-08-26 17:33:21
10	2016-12-19	2026-08-26 17:33:23	2026-08-26 17:33:34
11	2017-12-18	2026-08-26 17:33:36	2026-08-26 17:33:46
12	2018-12-24	2026-08-26 17:33:48	2026-08-26 17:33:58
13	2019-12-23	2026-08-26 17:34:00	2026-08-26 17:34:10
14	2020-12-21	2026-08-26 17:34:12	2026-08-26 17:34:22
15	2021-12-20	2026-08-26 17:34:24	2026-08-26 17:34:34
16	2022-12-19	2026-08-26 17:34:37	2026-08-26 17:34:47
17	2023-12-18	2026-08-26 17:34:49	2026-08-26 17:34:59
18	2024-12-23	2026-08-26 17:35:01	2026-08-26 17:35:11

Appendix B Per-step diagnostics

Realized in the projection tables below is the risk engine scoring its own forecast: the buy-and-hold return of the segment that followed each rebalance, on the same gross basis the cone was projected on. It is deliberately not the charged, calendar-window total return the study’s tables print, so the two will not reconcile line by line; the cone and its outcome share one basis, which is what a calibration test requires. Each row names its segment’s span so a boundary session is visible.

What each step's run actually did beyond its return: capital allocation across lanes and regimes, the portfolio book's rebalancing and cost drag, and how positions were sized. Harvested from the frozen run reports, present where the circuit produced them.

Open the full per-step grid (18 steps: every rebalance, capital routing and sizing, per window)

Step 1 · 2007-12-24 → 2008-12-23

Additions

Portfolio book, rebalanced annual · 2 constructions · 4 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 1 name dropped at load (4 names actually held across the window), weights renormalised onto the rest

Projection accuracy, realized outcome fell inside the P5–P95 cone in 50.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 20.32% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2007-12-24 → 2008-01-01	-4.4283%	1.0904%	6.8227%	-3.5607%	yes	2.6339%	0 / 4
2008-01-01 → window end	7.9578%	65.8089%	153.7624%	-38.1842%	no	2.587%	51 / 247

Deletions

Portfolio book, rebalanced annual · 2 constructions · 3 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 2 names dropped at load (3 names actually held across the window), weights renormalised onto the rest

Projection accuracy, realized outcome fell inside the P5–P95 cone in 100.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 16.73% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2007-12-24 → 2008-01-01	-4.7337%	-0.026%	4.8258%	-1.4341%	yes	2.4725%	0 / 4
2008-01-01 → window end	-39.7241%	-12.4281%	26.8382%	-22.403%	yes	2.4782%	42 / 247

Step 2 · 2008-12-22 → 2009-12-22

Additions

Portfolio book, rebalanced annual · 2 constructions · 8 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 3 names dropped at load (8 names actually held across the window), weights renormalised onto the rest

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2008-12-22 → 2009-01-01	-8.1118%	0.1785%	10.0075%	3.3223%	yes	3.7774%	0 / 6
2009-01-01 → window end	-34.5027%	17.7985%	114.796%	37.6007%	yes	3.7087%	4 / 245

Deletions

Portfolio book, rebalanced annual · 2 constructions · 5 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 6 names dropped at load (5 names actually held across the window), weights renormalised onto the rest

Projection accuracy, realized outcome fell inside the P5–P95 cone in 50.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 6.37% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2008-12-22 → 2009-01-01	-12.378%	-2.0008%	10.6345%	2.1625%	yes	4.3895%	0 / 6
2009-01-01 → window end	-77.4831%	-51.5122%	6.306%	205.8081%	no	4.3921%	16 / 245

Step 3 · 2009-12-21 → 2010-12-21

Additions

Portfolio book, rebalanced annual · 2 constructions · 4 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 3 names dropped at load (4 names actually held across the window), weights renormalised onto the rest

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2009-12-21 → 2010-01-01	-7.9114%	-0.0915%	9.2327%	0.0155%	yes	3.1274%	0 / 7
2010-01-01 → window end	-40.1696%	1.2069%	66.1895%	10.634%	yes	3.1443%	3 / 244

Deletions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 1 name dropped at load (6 names actually held across the window), weights renormalised onto the rest

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2009-12-21 → 2010-01-01	-11.1787%	-0.4799%	12.7155%	1.5994%	yes	4.6021%	0 / 7
2010-01-01 → window end	-56.6231%	-9.6382%	80.596%	40.6451%	yes	4.6041%	0 / 244

Step 4 · 2010-12-20 → 2011-12-20

Additions

Portfolio book, rebalanced annual · 2 constructions · 5 names held · selection: hold · 0.0% in cash · turnover 0.8× · cost drag 0.08%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2010-12-20 → 2011-01-01	-9.25%	1.7717%	13.9777%	0.0%	yes	3.8332%
2011-01-01 → window end	-2.0606%	81.2423%	223.9394%	-27.7255%	no	3.6117%	10 / 158

Deletions

Portfolio book, rebalanced annual · 2 constructions · 5 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 1 name dropped at load (5 names actually held across the window), weights renormalised onto the rest

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2010-12-20 → 2011-01-01	-9.5668%	0.3085%	11.1262%	0.9404%	yes	3.8462%	0 / 8
2011-01-01 → window end	-31.3549%	20.7083%	105.5981%	-9.9696%	yes	3.7465%	7 / 244

Step 5 · 2011-12-19 → 2012-12-18

Additions

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2011-12-19 → 2012-01-01	-8.0435%	0.8619%	10.5092%	2.7978%	yes	2.866%	1 / 8
2012-01-01 → window end	-12.745%	46.8169%	141.948%	13.1005%	yes	2.8671%	11 / 241

Deletions

Portfolio book, rebalanced annual · 2 constructions · 5 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2011-12-19 → 2012-01-01	-8.0293%	-0.6001%	7.3308%	7.0355%	yes	2.8473%	0 / 8
2012-01-01 → window end	-38.2143%	-4.8735%	43.952%	2.1303%	yes	2.8339%	5 / 241

Step 6 · 2012-12-24 → 2013-12-24

Additions

Portfolio book, rebalanced annual · 2 constructions · 9 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 2 names dropped at load (9 names actually held across the window), weights renormalised onto the rest

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2012-12-24 → 2013-01-01	-3.6107%	0.689%	5.1002%	0.3688%	yes	1.9809%	0 / 4
2013-01-01 → window end	-0.624%	39.7692%	96.0297%	39.27%	yes	1.9842%	5 / 247

Deletions

Portfolio book, rebalanced annual · 2 constructions · 10 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 1 name dropped at load (10 names actually held across the window), weights renormalised onto the rest

Projection accuracy, realized outcome fell inside the P5–P95 cone in 50.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 0.0% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2012-12-24 → 2013-01-01	-5.7353%	0.0206%	6.018%	0.7381%	yes	2.892%	0 / 4
2013-01-01 → window end	-43.6745%	-10.5037%	41.6605%	61.5788%	no	2.8926%	0 / 247

Step 7 · 2013-12-23 → 2014-12-23

Additions

Portfolio book, rebalanced annual · 2 constructions · 5 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2013-12-23 → 2014-01-01	-3.9873%	1.1145%	6.5145%	1.6923%	yes	2.0779%	0 / 5
2014-01-01 → window end	17.1703%	69.4778%	139.5659%	32.0987%	yes	2.0736%	20 / 246

Deletions

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2013-12-23 → 2014-01-01	-5.37%	0.2214%	6.173%	0.0015%	yes	2.2585%	0 / 5
2014-01-01 → window end	-28.2737%	6.8713%	55.3308%	2.3387%	yes	2.1691%	7 / 246

Step 8 · 2014-12-22 → 2015-12-22

Additions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 2 names dropped at load (6 names actually held across the window), weights renormalised onto the rest

Projection accuracy, realized outcome fell inside the P5–P95 cone in 50.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 7.17% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2014-12-22 → 2015-01-01	-3.3595%	0.9255%	5.783%	-0.4462%	yes	1.7293%	0 / 6
2015-01-01 → window end	12.2317%	50.3752%	102.8675%	-6.6125%	no	1.7293%	18 / 245

Deletions

Portfolio book, rebalanced annual · 2 constructions · 3 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2014-12-22 → 2015-01-01	-4.9465%	0.3155%	6.3463%	-1.9514%	yes	2.061%	0 / 6
2015-01-01 → window end	-17.4526%	17.8244%	69.5845%	12.1447%	yes	1.9955%	15 / 245

Step 9 · 2015-12-21 → 2016-12-20

Additions

Portfolio book, rebalanced annual · 2 constructions · 9 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 1 name dropped at load (9 names actually held across the window), weights renormalised onto the rest

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2015-12-21 → 2016-01-01	-4.3634%	0.5973%	6.3233%	0.4095%	yes	2.021%	0 / 7
2016-01-01 → window end	-8.691%	26.3574%	71.6796%	-15.0088%	no	1.999%	18 / 244

Deletions

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2015-12-21 → 2016-01-01	-5.3736%	-0.4578%	5.2169%	1.7082%	yes	2.2119%	0 / 7
2016-01-01 → window end	-36.5385%	-12.0067%	19.7765%	3.42%	yes	2.2142%	8 / 244

Step 10 · 2016-12-19 → 2017-12-19

Additions

Portfolio book, rebalanced annual · 2 constructions · 4 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2016-12-19 → 2017-01-01	-4.2694%	0.5921%	5.6388%	-1.7756%	yes	1.8001%	0 / 8
2017-01-01 → window end	-8.0548%	21.0054%	58.5148%	26.435%	yes	1.8003%	10 / 243

Deletions

Portfolio book, rebalanced annual · 2 constructions · 3 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1% · 1 name dropped at load (3 names actually held across the window), weights renormalised onto the rest

Projection accuracy, realized outcome fell inside the P5–P95 cone in 50.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 1.99% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2016-12-19 → 2017-01-01	-5.7964%	-0.3663%	5.3072%	-3.1024%	yes	1.7291%	0 / 8
2017-01-01 → window end	-34.4252%	-10.5562%	21.3645%	28.2878%	no	1.7301%	5 / 243

Step 11 · 2017-12-18 → 2018-12-18

Additions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2017-12-18 → 2018-01-01	-4.325%	0.7534%	6.0367%	-1.1637%	yes	1.9641%	0 / 8
2018-01-01 → window end	-5.1131%	26.4921%	66.5017%	9.7549%	yes	1.966%	23 / 242

Deletions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2017-12-18 → 2018-01-01	-6.6981%	-0.4011%	6.239%	1.9218%	yes	2.4291%	0 / 8
2018-01-01 → window end	-35.857%	-7.5447%	31.1347%	4.6938%	yes	2.4345%	7 / 242

Step 12 · 2018-12-24 → 2019-12-24

Additions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2018-12-24 → 2019-01-01	-3.5493%	0.5799%	4.8089%	7.1528%	no	2.0077%	0 / 4
2019-01-01 → window end	-1.1069%	37.8101%	91.5178%	48.6239%	yes	2.0079%	16 / 247

Deletions

Projection accuracy, realized outcome fell inside the P5–P95 cone in 50.0% of 2 scored rebalances (a well-calibrated 90% band contains ~90.0%) · 95% VaR breached on 4.38% of 251 days (expected ~5.0%)

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2018-12-24 → 2019-01-01	-3.6369%	0.2316%	4.1836%	4.8512%	no	1.6019%	0 / 4
2019-01-01 → window end	-17.4504%	12.5203%	52.9815%	1.1408%	yes	1.604%	11 / 247

Step 13 · 2019-12-23 → 2020-12-22

Additions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2019-12-23 → 2020-01-01	-4.2763%	0.809%	6.1913%	0.0541%	yes	2.3285%	0 / 5
2020-01-01 → window end	-0.5719%	43.9552%	103.6741%	37.0132%	yes	2.329%	26 / 246

Deletions

Portfolio book, rebalanced annual · 2 constructions · 6 names held · selection: hold · 0.0% in cash · turnover 1.0× · cost drag 0.1%

Rebalance	P5	Median	P95	Realized	In band	VaR 95 (1d)	Breaches
2019-12-23 → 2020-01-01	-4.7613%	0.0418%	5.1125%	0.0686%	yes	2.1117%	0 / 5
2020-01-01 → window end	-28.514%	1.6851%	41.5021%	-0.4572%	yes	2.1143%	34 / 246