The window everyone posts is the one where it worked. Stocks and long Treasuries
moved opposite ways in about a third of the rolling three-year windows since 1953. The
2010s are the most negative decade in the data. Build a case on that decade and you have
built it on the exception.
This model cannot see the thing that would actually kill you. UPRO and TMF reset
daily, so what ends a 3x fund is one
day down more than a third, not one month.
Everything here is monthly, which means it quietly assumes that day never arrives.
The legs are stand-ins. UPRO tracks the S&P 500 and this uses the US total
market. TMF tracks the ICE 20+ Year index and this uses a modelled rolling par 20-year
Treasury, which for the years Treasury issued no 20-year is really the 30-year. Fund
fees here are the 0.90% the rest of the site uses; UPRO and TMF both charge more.
Quarterly rebalancing is the easy part to write down. Every number above assumes
you sold the leg that went up and bought the leg that was collapsing, on schedule, for
decades, including in the quarters where both legs fell. That is the part people do not
do.
None of this says the strategy is wrong. It says the evidence for it is one favourable
stretch, and the page you are reading is the rest of the record. The
leverage lab will run a mix you choose with the financing
costs broken out, and
what leverage costs covers why the
multiple that wins is lower than it looks.