TMF trading strategies, backtested

Direxion Daily 20+ Year Treasury Bull 3X Shares: 3x daily leveraged long-term Treasuries, a leveraged rate bet. Every DeployQuant template run on TMF over 5.7 years of minute data, same engine, same window, sorted by return.

TMF is the Direxion Daily 20+ Year Treasury Bull 3X Shares, a fund that aims for three times the daily move of long-term US Treasuries. From 2021-01-04 to 2026-10-02 it lost 91.44% of its value, a CAGR of 34.82% negative, with annualized volatility of 45.67%. The fund never made a new high after the first day of the window, so its max drawdown of 91.44% and its total loss are the same number.

Buy-and-hold in the strategy engine, starting from $10,000 with no fees, ended at $1,172, a CAGR of 31.16% negative and a max drawdown of 88.49%. No rule set ended with a positive CAGR. Eleven of the twelve beat buy-and-hold on CAGR, and eleven had a shallower drawdown. The ranking on this page measures which rules lost less. It does not describe rules that made money.

The best result was the RSI(2) snapback at 4.0% negative, followed by the golden cross at 8.36% negative and the 200-day regime filter at 10.5% negative. The worst were the monthly cycle at 32.21% negative and the dip buyer at 29.84% negative, which are the two rules that stayed invested almost all the time. The pattern is plain: on a fund that lost value in each of the first five calendar years, time spent in the fund was the main cost, and the rules with the least exposure lost the least.

The trailing-stop templates, the momentum breakout and the trend plus trailing stop, are backtest-only for now, so their rows describe simulated history. All runs are daily-decision rules on minute data with no margin, and the headline run carries no fees or slippage.

Ranking the twelve rule sets on a fund that kept falling

The RSI(2) snapback made 170 round trips and won 56% of them. The average win was 3.22% and the average loss was 4.11%, which gave a profit factor of 0.95, just below break-even. Its Sharpe ratio was 0.01 and its max drawdown was 55.1%, which ran from 2022-01-28 to 2022-11-07 with no recovery in the window. It beat holding in all six calendar years, as did the RSI mean reversion. Its yearly returns were 16.8% in 2021, 45.4% negative in 2022, 30.1% in 2023, 22.7% negative in 2024, 44.2% in 2025 and 14.5% negative in 2026. A short-horizon rule can pick up the sharp rebounds a volatile fund produces, and TMF produced many of those.

The golden cross made 4 round trips and lost on all of them, each by an average of 12.05%, which is a 0% win rate and a profit factor of 0. Exposure was only 17.2%. It still lost 8.36% a year because each of its 4 positions closed at a loss. The 200-day regime filter made 17 round trips and lost on every one, with an average loss of 3.74% and exposure of 14.1%. Its nine trades in 2024 all lost.

The SMA 10/50 trend won 22% of its 18 trades, with an average win of 7.79% against an average loss of 6.6% and a profit factor of 0.34. The EMA 12/26 trend lost 13.9% a year on 26 trades and a 27% win rate. The 3-month momentum won 18% of 17 trades. The momentum breakout lost 11.1% a year on 27 round trips, with a 26% win rate and a profit factor of 0.56, though it had one strong year, 21.4% in 2023. The trend plus trailing stop won 22% of 9 trades and had a profit factor of 0.12.

The weekly 7% target made 212 round trips, more than any other rule, and lost 11.83% a year. Its average win was 7.34% and its average loss was 3.81%, with a 33% win rate and a profit factor of 0.85. It was invested 68% of the time and its drawdown of 60.65% began on 2021-01-04 and never recovered in the window.

The RSI mean reversion is the case where a high-looking number hides the damage. It won 46% of 13 round trips, but its average loss was 20.86% against an average win of 8.83%, a profit factor of 0.19. One trade, entered 2022-01-04 and exited 2022-11-25, lost 66.74%. Its max drawdown was 81.12%, close to holding, and it beat buy-and-hold by only 6.12 points of CAGR.

The dip buyer shows 0 closed round trips and exposure of 98.6%. It bought once, the profit target was never reached, and it held to the end, losing 29.84% a year against 31.16% for holding. The monthly cycle lost 32.21% a year, worse than holding, with 69 round trips and a 39% win rate.

The only profitable years were short rallies

Holding TMF lost money in every calendar year of the template run: 18.8% in 2021, 69.7% in 2022, 9.9% in 2023, 30.7% in 2024, 2.3% in 2025 and 23.1% in 2026 to date. The templates that made money in a year did so in 2023 and 2025, and both years were short rallies.

The 2023 rally came in the last two months of the year. Holding made 23.37% in November 2023, the best month of the window. The monthly cycle made 28.17% on its November trade and 26.69% on its December trade. The weekly 7% target made 31.3% in 2023, with 40.65% in November as its best month. The RSI(2) snapback made 30.1% and the SMA 10/50 29.2%, with December 2023 at 25.25% as its best month. The momentum breakout made 21.4% and the EMA 12/26 14.2%. Those two had their best trades in the same window, 28.48% from 2023-11-15 to 2024-01-03 and 13.9% from 2023-11-15 to 2024-01-19.

The same trades also explain the next year. The trend templates held into January 2024, when the fund fell again. The EMA 12/26 and SMA 10/50 both had January 2024 as their worst month, at -15.88% and -15.53%, and finished 2024 at -34.3% and -24%.

The 2025 column splits the field. The RSI(2) snapback made 44.2% that year against -2.3% for holding, and the weekly target made 6.9%. Every trend template lost, from -4% for the 200-day filter to -24.4% for the momentum breakout. The April 2025 shock cost the EMA 12/26 14.91% from 2025-04-02 to 2025-04-10 and the 3-month momentum switch 16.63%, and the weekly target 14.46% from 2025-04-07 to 2025-04-10.

Two open positions sit behind the weakest results. The dip buyer bought on 2021-02-02 at 260.51 and was down 90.18% at the end, so its 98.6% exposure is a buy-and-hold with a different entry day. The RSI(14) template entered on 2026-07-13 at 32.66 and was down 21.69%, on top of a closed trade that lost 66.74% from 2022-01-04 to 2022-11-25. Together these produced the -25.04% CAGR and 81.12% drawdown.

How each strategy traded TMF

Trades that mattered and what costs did

On a falling fund, the best trades are short rebounds. The RSI(2) snapback gained 12.83% from 2023-12-11 to 2023-12-14 and 10.09% from 2023-10-31 to 2023-11-02. The weekly target gained 11.95% from 2023-10-30 to 2023-11-02. The monthly cycle's two best trades, 28.17% in November 2023 and 26.69% in December 2023, came in the same stretch. The momentum breakout's best trade, 28.48%, ran from 2023-11-15 to 2024-01-03. That rebound in late 2023 is the one stretch in the window where long exposure paid off, and the month-of-year table points the same way: November has an average return of 12.46% in the data, against negative averages in most other months.

The worst trades are the ones where a rule held on. The RSI(2) snapback lost 16.03% from 2023-09-15 to 2023-10-02 and 13.36% from 2022-04-06 to 2022-04-21. The weekly target lost 14.46% from 2025-04-07 to 2025-04-10 and 13.91% from 2024-08-05 to 2024-08-08. Those losses are bigger than the 7% target and show how a weekly rule with no stop absorbs the fund's large daily moves. The monthly cycle lost 23.54% in April 2022 and 20.98% in October 2022.

The median trade tells the same story. The weekly target's median was a loss of 1.56% with a median hold of 3 days. The momentum breakout's median was a loss of 5.54% over 15 days. The trend rules' medians were losses of 3.45% to 12.7%. Only the RSI(2) snapback had a positive median trade, a gain of 0.72% over 4 days.

Costs add a little to each loss. The RSI(2) snapback went from 4.0% negative to 6.75% negative at 5 basis points and 9.46% negative at 10. The weekly target went from 11.83% negative to 13.91% and 15.53%. With 170 and 212 round trips, these two rules pay the most. The slow rules barely changed: the golden cross moved from 8.36% negative to 8.43% and 8.48%, and the dip buyer from 29.84% to 29.88% and 29.93%. The lead of the RSI(2) snapback over every other rule survives a 10 basis point cost, since its 9.46% loss is still smaller than the 11.04% of the 200-day regime filter at the same cost.

How TMF behaved

Calendar years

Biggest single days

Average return by calendar month

Most and least correlated funds

Leverage and decay against TLT

The last column compounds 3× TLT's daily return with no fees or financing: the return a perfect daily-reset fund would have had. It leaves out the fund's fees, its borrowing costs and the interest it earns on cash, which put a real fund below or above this line. A daily-reset fund does not aim for 3× the underlying's return over a year.

Over 2021-01-04 to 2026-10-02, TMF returned −91.4% while TLT returned −40.5% and a perfect daily-reset 3× version would have returned −85.9%. Its measured daily beta to TLT was 2.98.

Why TMF behaves as it does

TMF closed up on 48.51% of sessions. The average up day was 2.18% and the average down day was 2.33% negative. The average intraday range was 2.81%, and the average overnight gap was 1.63%. In 62 separate cases the fund fell 10% within 20 days, covering 526 trading days. The best day was 2022-11-10 at 11.57% and the worst was 2022-03-02 at 10.12% negative. The next worst days were 2022-06-13 at 9.23% negative, 2025-04-07 at 9.21% negative and 2023-05-01 at 8.47% negative.

Calendar-year returns were 19.44% negative in 2021, 72.63% negative in 2022, 11.4% negative in 2023, 35.98% negative in 2024, 2.89% negative in 2025 and 29.49% negative in 2026 for the part of the year in the window. There was no positive calendar year. The longest drawdown lasted 1,443 sessions.

The fund's tie to long Treasuries is nearly exact. Its correlation to TLT is 1.00 and its realized beta to TLT is 2.98, which matches the three-times target. Its correlation to SPY is 0.08 and to QQQ is 0.09, with betas of 0.21 and 0.18. TMF is a rates position and not an equity position. Its closest funds by correlation are TLT at 1.00, BND at 0.92, AGG at 0.91 and IEF at 0.91, and UST at 0.86.

The decay table shows what leverage did. Over the common window TLT itself lost 40.49% in total. A daily-rebalanced 3x position in TLT would have lost 85.94%. TMF lost 91.44%, a gap of 30.03 percentage points from three times the underlying move of 121.46% negative, a figure that cannot be reached as a loss since a fund cannot lose more than all of it. The per-year figures show the same direction. In 2022 TLT lost 31.21%, three times that would be 93.62% negative, and TMF lost 72.63%. In 2023 TLT gained 2.79% while TMF lost 11.4%, and in 2025 TLT gained 4.23% while TMF lost 2.89%. In both of those years a positive year for the underlying produced a negative year for the fund.

The month-of-year averages are lopsided. November averaged 12.46% and June 3.49%, and the remaining months were mostly negative, with September at 9.63% negative, October at 8.59% negative and April at 8.15% negative. Each month has only 5 or 6 observations, so these are descriptions of the window and not seasonal rules.

The split between overnight and intraday returns is unusual. The overnight log return was 260.41% negative and the intraday log return was 16.32% positive, so the overnight share of the return was 106.69% and the intraday share was 6.69% negative. Every point of the fund's loss came from holding it between the close and the next open, and the trading day itself was a small net gain. A rule that exits before the close and re-enters the next morning avoids the overnight leg, and the rules with the shortest holds, the weekly target at 5.7 days and the RSI(2) snapback at 5.5 days, were the two busiest rules, and the RSI(2) snapback lost least of all twelve. The monthly cycle, invested 95.2% of the time, took nearly every overnight gap.

On oversold readings, RSI(14) below 30 occurred on 75 sessions and the median forward 20-day return was 4.97% negative, against a baseline of 3.02% negative. RSI(2) below 10 occurred on 215 sessions with a median forward 5-day return of 0.29% negative, against a baseline of 0.7% negative. The short reading did slightly better than the baseline at 5 days. That fits the result: the fast rule lost least, and the slow mean-reversion rule, which waits for RSI(14) to fall and holds for weeks, lost 25.04% a year. The fund was above its 200-day average on only 16.39% of sessions and crossed it 33 times.

Among the other leveraged funds, the same RSI(2) snapback was the best rule on TQQQ at 39.31%, QLD at 26.17%, SSO at 24.56% and SPUU at 24.12%. SOXL was best with the golden cross at 45.93%, TECL with the monthly cycle at 35.97%, ROM with the weekly target at 30.27% and FAS with the dip buyer at 24.15%. Those funds had buy-and-hold CAGRs of 18.34% to 38.16%, while UST, the Treasury fund in the group, had 8.51% negative and its best rule, the SMA 10/50 trend, returned 0.68%. A rule set that suits an equity-index fund does not carry over to a leveraged rates fund, and TMF shows it more clearly than any other fund in the group: the RSI(2) snapback that earned 39.31% on TQQQ lost 4.0% a year here.

Falling episodes, weekdays and liquidity

The fund had 62 falls of 10% or more from a 20-day high, spread over 526 sessions. A dip-buying rule reads each of those as an entry. The cases where a template acted on it ended badly: the dip buyer bought once at 260.51 and the RSI(14) template bought into a 66.74% loss in 2022. The forward numbers explain why. After RSI(14) fell below 30, on 75 sessions, the median 20-day return was -4.97%, against -3.02% for all sessions. After RSI(2) fell below 10, on 215 sessions, the 5-day median was -0.29% against -0.7% for all sessions, and the 20-day median was -2.78%. After RSI(14) readings the 20-day result was worse than the average day, and after RSI(2) readings both medians were a little better than the average day, though still negative.

The weekday averages were -0.59% on Monday, 0.03% on Tuesday, 0.33% on Wednesday, -0.27% on Thursday and -0.2% on Friday. With a 2.33% average down day and a 2.81% average range, none of these is distinguishable from noise. The lag-1 autocorrelation was -0.04.

TMF traded about $162,498,054 a day, with a median minute volume of 3,160 shares. The cost runs at 5 and 10 basis points matter more for the 170 and 212 round trips of the RSI(2) snapback and weekly templates than for the rules with a few trades. At 10 basis points the snapback fell to -9.46% a year and the weekly target to -15.53%.

Correlations with the funds in the test are tight at the top and thin elsewhere. TLT sat at 1, BND at 0.92, AGG and IEF at 0.91 and UST at 0.86. TBF was at -1, with RINF at -0.4 and KMLM at -0.39. The fund's correlation to SPY was 0.08. The unleveraged comparison is TLT. On TMF the RSI(2) snapback made -4% a year against -31.16% for holding, so it finished ahead of the fund and still lost money.

Other leveraged etfs

Backtests are hypothetical, computed by DeployQuant's engine on minute-resolution consolidated US market data (2021-01-04 to 2026-10-02, $10,000 starting capital, no margin, no fees or slippage in the headline run; buy-and-hold puts 98% of the account in at the first open, as the templates do) and do not guarantee future results. Nothing on this page is investment advice. Live trading involves risk of loss.