Emerging markets worked surprisingly well here. Much better than I expected for 1976-2018.
Portfolio 1 is 30% UPRO, 30% TMF, 40% VWO.
Portfolio 2 is 30% UPRO, 40% EDV, 30% VWO.
Portfolio 3 is 40% UPRO, 60% TMF.
Portfolio 4 is 40% UPRO, 60% EDV.
Portfolio 5 is 40% UPRO, 60% IEF.
In the rough early part of that period, when the US dollar experienced high inflation & high interest rates & rising yields, from 1976-1981, only portfolio 5 posted positive nominal returns (7.49% CAGR) but the first two weren't that far negative (-0.73% and -0.27%) compared to the others, which had CAGR of -7.55% and -3.2%. Or, in simpler terms, the IEF portfolio had $15,400 on $10k while the others had $9572, $9840, $6243, and $8229. This also speaks relatively in favor of 60% EDV, which avoids a lot of the interest (-120% cash position), expenses, and duration exposure compared to 60% TMF.
From 1976-1995, the 30% UPRO / 30% TMF / 40% VWO portfolio had 15.41% CAGR with 0.52 Sharpe, 1.76 Sortino.
From 1976-1995, the 40% UPRO / 60% EDV had 14.04% CAGR with 0.44 Sharpe, 1.59 Sortino.
From 1996-2015, the 30% UPRO / 30% TMF / 40% VWO portfolio had 14.94% CAGR with 0.74 Sharpe, 3.14 Sortino.
From 1996-2015, the 40% UPRO / 60% EDV had 13.88% CAGR with 0.66 Sharpe, 2.51 Sortino.
There may be other opportunities to use 'risky' asset classes with high returns and relatively low correlation to UPRO in order to improve risk-adjusted return in a backtest and to improve the odds, in a very real sense, of getting higher returns in the future. Using UPRO and TMF makes room for these other asset classes by being efficient at getting exposure to their risks with a minimum of cash.
The two potential weak points of a UPRO/TMF portfolio are, in simple terms, UPRO and TMF. We don't usually think of the S&P 500 as being something that could drag down returns, but that wasn't much consolation in 2000-2003 or (say) to the buyers of Japanese stock in the late 80s. Apparently both of those weak points can be shored up by reducing the allocation to each and making room for other return-generating assets that aren't fully correlated with the S&P 500 or to long term treasuries.
P1 uses 30% UPRO / 30% TMF / 40% VWO (emerging markets).
P2 uses 30% UPRO / 30% TMF / 40% VEA (developed markets).
P3 uses 30% UPRO / 30% TMF / 40% VNQ (real estate).
P4 uses 30% UPRO / 30% TMF / 40% VWEAX (high yield bonds).
P5 uses 30% UPRO / 60% TMF.
Given that the falling rates were obviously great for backtesting the CAGR of 60% TMF, and the current low rates (relative to a 1979 starting point) represent a headwind for 60% TMF (since they can't harvest the same degree
of benefits of falling rates), these alternatives do look reasonable. Diversifying further with that extra 40% of space in the portfolio may be even more reasonable.
* The 1976 and 1979 starting points were based on the youngest fund / source of data. The 2015 end point, the second time, was because I forgot to edit the cell back to 2018... but I suppose it's interesting nonetheless.