A harder test for my statistical portfolio

My dissatisfaction with the statistical portfolio has become more specific. Technology shares have produced extraordinary gains over the past several years. A portfolio that constantly ranks and replaces shares should have to explain what that activity adds to simply owning the businesses driving those gains.

The first reassessment concentrated on reducing turnover and controlling losses. That produced two useful experiments: slower stock momentum and allocation across asset classes. But a cautious allocation portfolio answers a different question from an investor looking for substantial long-term growth. I had not given a simple growth investment a demanding enough place in the comparison.

I have now tested that comparison, including Treasuries. The results give me a stronger candidate to investigate and a much clearer reason to keep a simple alternative in view.

The table below starts each portfolio with $100,000 on 17 October 2016 and ends on 29 October 2026. Trading costs are assumed to be 0.10% of every purchase or sale, plus $1 per order. Fund expenses are already reflected in fund prices. Returns are nominal US dollars, before personal taxes and currency conversion. ETF distributions are reinvested; stock strategies reinvest available cash on their scheduled rebalances. The individual Treasury calculation keeps its coupons in cash.

Approach	Ending value of $100,000	Annualised return	Largest fall from a previous peak
Concentrated 20-stock momentum, exploratory	$2,323,660	36.8%	-36.5%
Semiconductor fund, SMH	$1,824,356	33.6%	-45.3%
Slower stock momentum, 100 positions	$719,290	21.7%	-33.8%
Nasdaq-100 buy-and-hold, QQQ	$653,080	20.6%	-35.1%
Simple diversified portfolio, 60% VT / 40% BND	$218,672	8.1%	-22.2%
Reconstructed original statistical rules	$200,585	7.2%	-36.0%
Allocation across asset classes	$182,061	6.2%	-8.4%
Individual Treasury note held to maturity, coupons kept in cash	$117,678	1.6%	Not measured
Rolling 7–10-year Treasury fund, IEF	$104,546	0.4%	-24.0%

These are research calculations. The original statistical row reconstructs the 100-position rules; it is not the actual history of the live account or the separate ten-stock Weekly Winners portfolio. The stock tests use the present configured universe, which makes them vulnerable to survivorship and selection bias. That limitation matters particularly for the strongest result in the table.

The Nasdaq-100 result is a useful rebuke to unnecessary activity. After the assumed costs, $100,000 became approximately $653,080. The slower 100-position strategy also performed well in this particular decade, slightly exceeding QQQ, although it fell behind over the longer period beginning in 2008. There is no reason to pretend that one investment wins in every window.

Semiconductors did considerably better still. The historical SMH result shows how profitable a sustained technology exposure could have been. It also reflects a narrower industry bet. Selecting that industry now, after its exceptional performance, does not establish that it was the obvious allocation before the gains occurred, or that the next decade will reward it in the same way.

There is also a clean, recent test of owning a handful of familiar technology companies. The fund now called MAGS began with five companies in April 2023 and added Nvidia and Tesla in October. From the first full session after that change, 3 October 2023, its annualised return in this calculation was 32.5%, versus 25.8% for QQQ, with larger losses along the way. That is a real investable fund history. A seven-stock basket projected back into 2013 would tell a different, much more flattering story about what investors supposedly knew. Roundhill's October 2023 announcement

The new stock strategy is straightforward. It holds twenty equity securities from the configured universe, ranks them on total returns over the previous twelve months excluding the latest month, and reviews the portfolio monthly. Existing positions can remain while they rank in the top thirty. It targets equal weights, keeps 2% in cash and limits each configured sector group to ten positions at rebalancing. Orders use the next session's opening price. These rules were fixed before the new results were calculated.

Its 36.8% annualised return over the past decade deserves attention. Raising the assumed trading friction from 0.10% to 0.25%, while retaining the $1 charge, reduced that result to 35.7%. A predeclared quarterly variant achieved almost the same growth over the full period with substantially fewer trades, so a monthly schedule has not earned automatic preference.

The longer test is sobering: from July 2008 its annualised return was 21.6%, and its largest decline was 61.4%. QQQ returned 17.3% annually over that same longer window and lost 47.0% at its worst. The longer comparison itself starts after the technology bubble, so even that does not capture the full historical risk of technology investing.

This twenty-position version also changes equity eligibility and sector limits. Its improvement cannot be attributed entirely to owning fewer securities. Twenty listed securities can include two share classes of one company, as happened with Alphabet; they do not necessarily represent twenty distinct businesses. Weights can drift between rebalances. This is a price-momentum strategy that can also select energy, healthcare or other industries when their shares lead the market.

Published research gives a reasonable basis for investigating concentrated momentum. Jack Vogel's portfolio-construction study found stronger gross historical returns with more focused portfolios, but examined fifty to five hundred stocks before trading costs. It did not validate this twenty-stock recipe. Actual implementation is a useful counterweight: Alpha Architect's QMOM reported a ten-year annualised NAV return of 11.92% through September 2026, compared with QQQ's 20.91% over the same period. They have different exposures, but the gap makes it difficult to claim that a momentum label alone provides an advantage. Vogel's study, QMOM's published results, Invesco's QQQ results

The bond comparison needs equal care. For the individual Treasury row I used an actual October 2016 auction: a 1.5% coupon note, bought below face value, with a quoted yield of 1.793% and maturity in August 2026. Including accrued interest and the same assumed acquisition friction, $100,000 produced about $117,678 in principal, coupons and leftover cash. That cash earned no further interest. This is a terminal-wealth calculation; it should not be confused with the cashflow yield. Official Treasury auction result

IEF continually maintains a portfolio of seven-to-ten-year Treasuries. Its shareholder has no fixed date at which the fund returns the original investment. Rising yields can depress its price for years. That explains why holding this fund for ten years produced a different outcome from buying one note and waiting for its maturity. I have not retrieved a daily price history for the individual note, so its interim drawdown is unknown rather than zero.

The starting yield also changes the forward decision. The Treasury's ten-year par yield was 4.98% on 29 October 2026, considerably above the 2016 purchase yield. That is a current market reference, not the promised return on every Treasury fund. A new growth strategy now has to justify its uncertainty against a much stronger bond alternative than the one available ten years ago. US Treasury yield curve

I also tested a slower trend rule on QQQ: own it above its ten-month moving average of monthly total-return values, otherwise hold short Treasuries through SHV. Across the longer 2008-onward test it reduced the largest decline to 28.6%, while annualised growth fell to 13.6%. It offers a visible trade-off between participation and protection. It does not preserve all the gains of buy-and-hold while removing the losses.

My plan is now narrower. Simple Nasdaq-100 ownership becomes the growth benchmark. The twenty-stock model is the candidate that must earn a place beside it. Before relying on its returns, I want a historical universe that includes the companies that failed, proper treatment of corporate actions, and independent periods that were not used to design the rules. A forward research ledger can check execution, costs and accounting over twelve monthly decisions; twelve months would not establish a durable investment edge.

Treasuries belong in the decision according to when money must be available and in which currency. A contractual dollar payment in ten years is useful for a dollar obligation. It provides no fixed purchasing-power guarantee in another currency. The amount reserved for that purpose should be chosen before deciding how much equity risk to take.

The existing portfolios remain unchanged. The next stage has a clear growth benchmark, a documented candidate, and specific questions that better data can answer.