Between the close on August 19 and the close on October 14, 2026, the iShares Semiconductor ETF (SOXX) went from $519.67 to $519.42. Thirty-nine trading sessions, a climb to $589 in early October and a slide back, and the chip index ended where it started. Over the same stretch the Direxion Daily Semiconductor Bull 3X (SOXL) lost about 8.4% and its mirror image, the Direxion Daily Semiconductor Bear 3X (SOXS), lost about 8.5%, both counting their September distributions. The bull fund and the bear fund both lost money on a flat index. Blame the arithmetic. It’s written into the prospectus of every daily-reset leveraged fund, and it shows up most clearly when a market goes sideways.
SOXL and SOXS both track three times the daily move, long and short, of the NYSE Semiconductor Index, the same index SOXX follows. ProShares Ultra Semiconductors (USD) offers two times the daily move of a different benchmark, the Dow Jones U.S. Semiconductors Index. The key word in all three is daily. Each fund resets its exposure every evening so that tomorrow it again holds 3x (or 2x, or minus 3x) of its current assets. What the fund promises is a multiple of one day’s return. It promises nothing about a week, a month or a year.
The two-day example everyone should run once
Start an index at 100. Day one it rises 10% to 110. Day two it falls 9.09% and lands back at 100. Flat. Now run the leveraged funds through the same two days.
| Fund | After day 1 | After day 2 | Same pattern repeated for 20 days |
|---|---|---|---|
| Index (1x) | 110.00 | 100.00 | 100.00 |
| 2x long | 120.00 | 98.18 | 83.24 |
| 3x long | 130.00 | 94.55 | 57.07 |
| 3x inverse | 70.00 | 89.09 | 31.50 |
The 3x long fund gains 30% on day one and then gives back 27.27% of a bigger base. It ends at 94.55. The inverse fund loses 30%, then gains 27.27% on a smaller base, and ends at 89.09. Repeat that up-and-back cycle ten times and the 3x long fund has lost 43% of its value while the index hasn’t moved; the inverse fund has lost more than two thirds. Nobody needs to be wrong about direction for this to happen. The index went nowhere.
Ten-percent days are extreme, so here is a gentler path. I built a 20-day sequence with daily moves between 1% and 4% in both directions (+3%, -4%, +2.5%, -2%, +3.5% and so on) that ends exactly flat. That works out to an annualised volatility of about 43%, which is ordinary for semiconductor stocks in 2026. After 20 sessions the index is at 100.00. The 2x fund is at 98.54. The 3x fund is at 95.66, and the 3x inverse fund is at 91.62. Four weeks of chop, no trend at all, and the bear fund is down more than 8%.
The formula behind the drag
The loss has a name, volatility decay or variance drag, and it has a tidy approximation. For a fund with daily leverage L, the drag over a period is roughly L(L-1)/2 times the variance of the index’s daily returns summed over that period. Expressed in log terms, the leveraged fund returns about L times the index’s log return, minus that drag.
Plug in the multipliers and the asymmetry shows itself. For a 2x fund the factor is 2×1/2 = 1. For a 3x fund it’s 3×2/2 = 3. For a minus 3x fund it’s (-3)×(-4)/2 = 6. So the 3x bull carries three times the drag of a 2x fund, and the 3x bear carries six times. Inverse leverage is the most expensive structure to hold through choppy markets, which is exactly what the table above shows.
The approximation is close enough to use. On my 20-day path it predicts 98.55 for the 2x fund, 95.72 for the 3x fund and 91.62 for the inverse fund, against exact values of 98.54, 95.66 and 91.62. Scaled up to a year, a flat index with 50% annual volatility would leave a 2x fund about 22% lower, a 3x fund about 53% lower and a minus 3x fund about 78% lower, before a dollar of fees. At 70% volatility the 3x bull loses about 77% in a year that goes nowhere. Volatility, squared, is the bill.
The real August to October window fits the same pattern. Using SOXX’s daily closes, the index ran at about 35% annualised volatility over those 39 sessions. A frictionless 3x fund compounding those exact daily moves would have lost about 5.9%; a frictionless minus 3x fund, about 10.7%; a 2x fund, about 2.0%. The actual funds landed in the same territory, with SOXL a little worse than the pure math and SOXS a little better. The gap is the costs covered below. The direction and rough size of the damage is the decay.
Compounding cuts both ways
None of this means leverage always loses. The same compounding that grinds a fund down in a range pays out in a trend. Take 20 straight sessions of +1%. The index gains 22.0%. A naive reading says the 3x fund should gain 66%; it actually gains 80.6%, because each day’s 3% is earned on a larger base. Flip it to 20 sessions of -1% and the index falls 18.2% while the 3x inverse fund gains 80.6%. In a smooth trend, leverage over-delivers. The opposite fund in each case loses 45.6%, which is still less than three times the index move, so compounding also softens the damage on the wrong side of a clean trend.
That’s the whole trade-off in one line. Daily leverage pays for persistence and charges for reversals. Semiconductor stocks in 2026 have offered both. The run into early October was close to a trend, and SOXL rose from about $101 on September 14 to $164 by October 5. The weeks around it were reversals stacked on reversals, including a 6.8% one-day drop in SOXX on October 22. Readers who followed the collapse of Korea’s single-stock leveraged chip ETFs this summer saw the violent version of the same mechanics: a 25% index drop turned into 70% losses once daily resets were layered on top.
The costs that sit on top of the math
Nobody collects variance drag; it simply leaks out of the fund. The funds also carry costs that somebody does collect. Direxion lists SOXL at a 0.75% net expense ratio (0.91% gross, with a fee waiver running to September 1, 2027) and SOXS at 1.00%. ProShares lists USD at 0.94%, with its own contractual waiver through September 30, 2027. On a flat year those numbers are small next to a 50% decay figure. They still add up for anyone holding through quarters.
The bigger hidden cost is financing. A 3x long fund holds about 300% exposure on 100% of capital, mostly through total return swaps with banks. ProShares’ own holdings list shows USD’s swap book spread across nine counterparties. The bank charges a floating short-term rate plus a spread on the borrowed portion, and that charge flows through the swap’s return; it never appears in the expense ratio. When short rates are a few percent, financing two extra units of exposure costs a 3x long fund several points a year. Inverse funds sit on the other side. They hold cash and Treasury bills as collateral and earn interest on them, which partly offsets their larger variance drag. That’s consistent with the August to October window, where SOXL trailed its frictionless 3x path and SOXS did slightly better than its own.
Add the pieces together and a leveraged semiconductor fund held through a sideways quarter faces three drains at once: variance drag that scales with the square of volatility, an expense ratio near 1%, and, for the long funds, a financing bill that rises with interest rates. Only the first one is large, and it is the one most holders never compute.
The practical reading is narrow. A leveraged chip fund is a bet on the path the index takes as much as on where it ends up. If you think semiconductors end the quarter higher but expect a bumpy ride, SOXL can lose money while you’re right. If you think they end lower, SOXS can do the same. Anyone tracking the sector through funds like these can measure the risk directly: take the index’s recent daily volatility, square it, multiply by L(L-1)/2 and by the holding period. That number is the toll. Analysis.org covers semiconductor stocks regularly, and the fund-level math applies to every one of them.
This post explains how daily-reset leveraged and inverse ETFs behave. It isn’t a recommendation to buy or sell any security.