Things we tested that did not earn a place

How to read these

A rejection is only as good as the series that produced it

Three things get called a null here. We never looked. We looked and our data was too thin to see an effect the size of the one we found. Or we found something real pointing the other way. Only the third is a result.

Here is a case where we got it wrong. We rejected corporate credit for moving almost in lockstep with long Treasuries. Then we checked what we had measured it on. The index carried twenty years of interest-rate risk, and that was most of what the correlation was picking up. Hedge that part out and the link to Treasuries is close to nothing, over nearly ninety years of monthly returns. We withdrew the rejection.

It happened again with foreign value funds. On the short window two of them impose, all four looked clearly bad. Widen it and they read −1.0 and −1.7 percentage points a year against the model, both ranges covering zero. So each section below says how broad its verdict is and what would reopen it.

Timing

Selling when the market drops below its own long-run average

Hold the index above its long-run average, sell to cash below it, buy back when it crosses over. We cannot tell you whether that works. On US data the rule gained +0.74 percentage points a year against a control taking the same average market risk, while the smallest gain that test could have detected was 3.03 percentage points a year. Our instrument could not see an effect that size. A far larger test, pooling sixteen countries across a century and a half, does find the effect and can see it.

What rules it out for a US taxable investor is narrower. The rule creates its own tax bill, worth about 1.92 percentage points a year against running it in a sheltered account, and cheaper things buy the same protection. Holding less stock buys it free. A trend fund buys the same signal across dozens of markets without selling your stocks.

One part survives. Cutting off the worst outcomes is arithmetic rather than a reward for risk, so the rule does cut the worst fall a long way: −43.1% against −71.6% for a control taking the same average market risk, and −83.7% for simply holding.

Crypto

Bitcoin as the thing that zigs when stocks zag

Parts of the case have improved. The spot funds hold real money, the price swings less wildly, and regulation opened up. None of that is the diversification claim, which is what we tested.

For every move in US stocks, bitcoin moves 1.53x up, 1.62x down. It falls harder than it rises, which is backwards for something you buy as ballast. Across the worst tenth of months for US stocks over the last decade, bitcoin was positive in 2 of 12, and one of those two was a bull market in bitcoin overlapping an equity dip. It also made the portfolio's worst loss deeper at every weight we tried. That is a real result, and a broad one. Returns did not rescue it either: −33.2% against +18.7% over the past year, and +56.3% against +72.8% over five.

Hold it if you want to. One or two per cent, labelled a speculation, is defensible, and so is zero. Past about five per cent it is a borrowed-money bet on stocks bought at a bad price.

Crash insurance

Funds that promise to pay off in a crash

The pitch is a fund that pays enormously when the market falls apart, so you can hold more stock the rest of the time. The payoff has to bend for that: modest help in an ordinary fall, huge help in a real crash. None of eight gave us a bend we could measure. Cambria's TAIL has compounded at −7.15% a year against SPY's +15.35% a year on the issuer's own factsheet since 2017, and Simplify's CYA fell −99.16% before being liquidated in 2024.

One caution about our own number. TAIL is nine-tenths ten-year Treasuries and a sliver of put options, so most of that loss is the worst bond market in forty years rather than the options decaying. Cash still beats it outright, adding back more in the average bad month than anything on the option shelf manages after its own costs.

Tax machinery

Direct indexing, and harvesting inside a plan that never sells

Instead of one index fund you own the individual stocks behind it, and a computer sells the ones that are down so you collect the tax losses. The fee is not why we said no. Set it to zero and it still trails owning the fund.

Losses only save you money if you have gains for them to cancel. A taxable account that never sells produces none, so the losses queue behind that $3,000 cap and only 0.2% of every dollar harvested ever produces a tax saving. The outside gains you would need to break even, as a share of the taxable account: 1.2% a year held to death, 3.0% if ever sold. It also doubles the cage it exists to exploit, since after thirty years the locked-in gain reads 55.7% of value against 26.7%, harvested against left alone.

The plain version survives, and you should still do it. Sell one total-market fund at a loss, buy a different sponsor's. No fee, reversible, and it wins below that break-even.

Three more

Gold, anti-beta funds, and three well-built factor products

Gold does diversify. Over half a century its return per unit of bumpiness scores 0.18 against 0.42 to 0.59 for stocks, and it moves with stocks at about zero rather than against them. Cash diversifies nearly as well and costs nothing. Both readings we took sit inside what the test could resolve, so gold is a shrug.

Funds that own the boring stocks and bet against the exciting ones are sold as the defensive move. The one bend we could measure anywhere on our panel is this one, and it points the wrong way: the strategy gets worse as stocks get worse, because a crisis is when borrowing is withdrawn. BTAL has returned −3.63% a year against the S&P 500's +15.41% a year since launch.

Two momentum funds and a deep-value fund lost on arithmetic that says little about the underlying idea. MTUM trades so much that the costs exceed the exposure bought. SPMO is the same idea built cheaper, and at a 5% weight it comes out at +0.02% a year against the same portfolio without it, because it overlaps heavily with the momentum fund we hold. RPV gets the deepest value exposure available by giving up momentum, inside a portfolio that holds momentum on purpose.

Rebalancing

Rebalancing as a source of return

Selling what has risen to buy what has lagged is widely sold as a free bonus. It does not pay, and that is a genuine negative result rather than a shrug. Measured against never trading at all the policy returned −0.09% a year, too small a difference for that test to read either way. A separate test on a different mix, over thirty-five years, put it further behind than that and could see the gap. The bonus needs the things you hold to drift apart at similar rates, and two real stock markets do not.

Keep rebalancing anyway. It is the cheapest way to hold your exposure where you said it would be. An annual review with a band around each target keeps the portfolio within 0.94 percentage points of target, against roughly six times that if you leave it alone.

The pattern

What these have in common

Two verdicts above are the strong kind. Bitcoin as ballast fails on its own measured shape, and the rebalancing bonus is measurably negative here. The rest are narrower than they look: a tax position, a product shelf, a portfolio that already owns the thing on offer, or a test too weak to read. Credit is why we say so. It sat on a page like this one for a year, wrongly.