UDN trading strategies, backtested

Invesco DB US Dollar Index Bearish Fund: profits when the US dollar weakens against major currencies. Every DeployQuant template run on UDN over 5.7 years of minute data, same engine, same window, sorted by return.

UDN is the Invesco DB US Dollar Index Bearish Fund. It rises when the US dollar weakens against a basket of major currencies and falls when the dollar strengthens. Each of the 12 DeployQuant templates started with $10,000 on 2021-01-04 and ran to 2026-10-02. Buying and holding UDN lost 1.12% a year, ended at $9,375, and had a 22.74% max drawdown that had not recovered by the last day. The Sharpe was negative 0.12.

All 12 templates beat holding on CAGR and all 12 had a shallower drawdown. Seven finished with a positive CAGR. The best is the trend plus trailing stop rule at 1.88% a year with an 8.28% drawdown, followed by the SMA 10/50 trend at 1.53% and the 200-day regime filter at 1.26%.

This result needs context. A fund that returned negative 1.12% a year with a 7.06% annualized volatility gives every rule a low bar. Beating holding here means losing less or earning a point or two, and the best CAGR of any template is under 2%. Two templates made no trades at all and finished at exactly $10,000, which also counts as beating a fund that lost money.

The figures are one backtest with daily-decision rules, fills on minute bars, no margin, and no fees or slippage in the headline run. Two cost runs add 5 and 10 basis points. The page describes what the templates did on this fund in this window. It does not forecast what they would do in a different one.

Why a flat, drifting fund favoured the rules that stayed out

UDN spent the window drifting. Its calendar years were negative 7.07% in 2021, negative 7.95% in 2022, 4.69% in 2023, negative 4.36% in 2024, 12.42% in 2025 and negative 2.36% in 2026 to date. Its price series lost 5.99% in total. The max drawdown of 23.24% ran from the peak on 2021-01-06 to the trough on 2022-09-27, and the fund never recovered that peak. The longest time under a prior high was 1,441 sessions out of 1,444.

In that setting the templates that were flat in 2021 and 2022 had a large advantage. The trend plus trailing stop rule returned 0% in 2021 and 1.5% in 2022 against negative 6.9% and negative 7.8% for holding. The 200-day filter returned 0% and 1.2%. The golden cross returned 0% in both years. The two templates that never traded, the dip buyer and the momentum breakout, returned 0% in every year. They beat holding in 2021, 2022, 2024 and 2026 by sitting still, and trailed it in 2023 and 2025, when the fund rose 4.6% and 12.1%.

The trend plus trailing stop rule is the strongest case. It had no completed round trips, yet it was invested 66.6% of the time. That combination means it entered once and held a single position through the end of the window. It made 11.31% in total with a drawdown of 8.28% that began on 2024-09-27, and it returned 12.2% in 2025, matching the fund's best year. The rule entered during 2022 and the stop never fired. One position is a small sample, and the result should be read as one trade that worked.

The SMA 10/50 trend rule made 17 round trips and won 35%, with an average win of 3.45% and an average loss of 1.01%, giving a profit factor of 1.82. It lost 2.3% in 2021, gained 3.7% in 2022 and 4.4% in 2023, and gained 9% in 2025. The 200-day filter made 20 round trips with a 50% win rate, a profit factor of 2.5 and the shallowest drawdown of any template that traded, at 5.38%. The EMA 12/26 trend rule made 22 round trips, won 23%, and lost 8 trades in a row at its worst, finishing at 0.99% a year.

The golden cross lost 1.06% a year, nearly the same as holding at negative 1.12%. It made 4 round trips and won 1. Its average hold was 264.2 days. It sat out 2021 and 2022 and then entered into 2023, where it lost 2.7%, and it missed most of the 2025 gain of 12.1%, taking 2.3% of it.

The mean reversion rules and the weekly target

The RSI mean reversion template made 0.62% a year with 17 round trips and a 65% win rate. Its average win was 1.45% and its average loss was 1.91%, so a high hit rate produced a small return, and the max drawdown of 14.61% ran from 2021-04-19 to 2022-09-27 and recovered only on 2026-01-26. The rule was invested 50.6% of the time. It lost 3.7% in 2021 and 4% in 2022, because the decline continued after its entries.

The fund's own statistics show a mild edge. The 14-day RSI closed under 30 on 61 sessions. After those the median 5-day forward return was 0.25% against a baseline of negative 0.05%, and the median 20-day return was 0.48% against negative 0.11%. The sample is larger than for most funds in this dataset, and the edge is small in absolute terms. The 2-day RSI closed under 10 on 200 sessions, with a median 5-day return of 0.03% and a median 20-day return of negative 0.3%, which is no better than the baseline.

That explains the RSI(2) snapback. It made 171 round trips, won 56%, and was invested 45.4% of the time. Its average win of 0.52% was smaller than its average loss of 0.61%, which gives a profit factor of 1.07, and its CAGR was 0.4%. Its max drawdown was 17.34%, from 2021-06-10 to 2022-09-27, and it lost 10.3% in 2022. This rule has an edge so thin that costs remove it.

The cost runs make the point. At 5 basis points the snapback's CAGR goes from 0.4% to negative 2.49%, with end equity of $8,651. At 10 basis points it goes to negative 5.31%, with end equity of $7,312, a Sharpe of negative 1.121 and a drawdown of 27.37%. At that level the rule is worse than holding. The average trade is 0.52% on wins, so a charge of 5 or 10 basis points on each side takes a large share of it.

The weekly 7% target is a stranger case. It was invested 91.4% of the time and made 62 round trips but won only 8% of them, and its longest loss streak was 31 trades. UDN has an average intraday range of 0.4%, so a 7% weekly target is out of reach for most weeks. The few trades that reached it paid an average of 7.45%, against an average loss of 0.6%, which produced a profit factor of 1.05 and a CAGR of 0.26%. The average hold was 29.8 days, since a trade that is not losing on Thursday carries on. Costs turned it negative: negative 0.59% at 5 basis points and negative 2.24% at 10.

The monthly cycle was invested 95.2% of the time, lost 0.67% a year, and had a 22.64% drawdown that matched holding almost exactly. It is the template closest to holding in both return and drawdown. The 3-month momentum rule lost 0.47% with 4 round trips and a 16.3% exposure.

How each strategy traded UDN

Trade statistics and what costs did

The trade table splits the templates into three groups. Two never traded. Three held positions for months: the golden cross for 264.2 days on average, the 3-month momentum rule for 86.2, and the SMA 10/50 trend for 55.5. The rest held for days to weeks, with the snapback at 5.5 days.

Costs changed the ranking. The trend and trailing-stop rule went from 1.88% to 1.87% at 10 basis points. The golden cross went from negative 1.06% to negative 1.19%. The slow templates barely moved. Among the more active ones, the SMA 10/50 trend fell from 1.53% to 0.94%, the 200-day filter from 1.26% to 0.57%, the EMA trend from 0.99% to 0.24%, and the RSI(14) template from 0.62% to 0.02%. By the 10 basis point run, four of these were within a point of zero.

The monthly cycle moved from negative 0.67% to negative 3%. The weekly target moved from 0.26% to negative 2.24%, and the snapback from 0.4% to negative 5.31%. In the 10 basis point run, five templates stayed positive: the trend and trailing-stop rule, the SMA 10/50 trend, the 200-day filter, the EMA trend and, at 0.02%, the RSI(14) template. The two templates that did not trade stay at 0%.

UDN is a thin fund. It traded an average of $2,009,419 a day in the window, with a median minute volume of 400 shares. That volume is small next to the funds on other pages. The cost runs charge a flat number of basis points per fill and do not model a wider spread on a thin fund, so the 10 basis point run is the one closer to a real small-account experience, and it suggests that frequent trading in UDN gave back everything it earned.

The worst trade for any template was negative 5.13%, for the RSI(14) template, and the best was 9.68%, for the SMA 10/50 trend. Those are small next to the leveraged funds, and they show the fund's low volatility in trade terms. The longest losing streaks were 31 for the weekly target, 9 for the snapback and 8 for the EMA trend, which are all high-frequency or low-win-rate designs.

How UDN behaved

Calendar years

Biggest single days

Average return by calendar month

Most and least correlated funds

What UDN did in the window

UDN returned negative 5.99% in total, or negative 1.07% a year, with an annualized volatility of 7.06%. The fund rose on 44.84% of sessions, with an average up day of 0.37% and an average down day of negative 0.35%. Its average intraday range was 0.4% and the average overnight gap was 0.25%. There were no drawdown events of 10% or more within 20 days.

The best days were 2.37% on 2022-11-10, 2.09% on 2025-04-10, 1.96% on 2022-11-04, 1.7% on 2025-04-03 and 1.6% on 2023-11-14. The worst were negative 1.59% on 2022-09-23, negative 1.42% on 2022-09-13, negative 1.36% on 2025-05-12, negative 1.31% on 2022-07-05 and negative 1.3% on 2023-03-07. The largest day in either direction is 2.37%, which is small next to the leveraged funds. The fund's trough on 2022-09-27 came shortly after its worst day, 2022-09-23, and its best day on 2022-11-10 came later that autumn.

The overnight and intraday log returns were both negative, at negative 4.68% and negative 1.87%, which gives an overnight share of 71.5% and an intraday share of 28.5%. The fund lost value in both parts of the session, and most of the loss came overnight.

First-order autocorrelation was 0. The fund closed above its 200-day average on 57.59% of sessions and crossed it 40 times. For a fund with such small daily moves, 40 crossings in 1,444 sessions is a lot, and it is why the faster trend rules, the SMA 10/50 and the EMA, made 17 and 22 round trips with low win rates of 35% and 23%. The 200-day rule itself made 20 round trips.

Beta to SPY was 0.12 and to QQQ 0.08, with correlations of 0.28 and 0.24. The fund has a mild positive relationship with stocks, so a weak dollar and rising equities showed up together in parts of the window. In April 2025 the fund's best month in the backtests was 4.49% for the trend rules, which fits the 2.09% day on 2025-04-10 and the 1.7% day on 2025-04-03.

Calendar months, correlations and currency peers

The average return by calendar month was highest in November at 1.11%, December at 0.72% and April at 0.56%. It was lowest in September at negative 1.32%, October at negative 0.88% and June at negative 0.76%. Each month has 5 or 6 observations, and 2022 and 2025 dominate. The averages describe this window and do not give a seasonal rule.

UDN's closest relative in the data is FXE, the Euro trust, with a correlation of 0.96. That fits a fund that profits from a weaker dollar. The most negatively correlated fund is USDU, the dollar-bullish fund, at negative 0.9. Other correlations are weak, with IAU at 0.44, EEM at 0.43 and IGIB at 0.41.

The currency peers show how much the direction of the dollar mattered. FXE returned negative 0.9% a year when held and its best template, the trend plus trailing stop rule, made 2.14%. USDU, the dollar-bullish fund, returned 5.22% when held, and its best template, the monthly cycle, made 5.4%. On UDN, holding lost 1.12% and the best template made 1.88%. On USDU a rule that stayed in the fund earned about what holding earned, and on UDN a rule that stayed out earned more than holding. The same trend and stop rule was the winner on both FXE and UDN.

That shows the pattern for currency funds in this window. Where the underlying trend was against the fund, flat rules won. Where it was with the fund, staying in won. The templates do not know which case they are in, and the results come from which one the window happened to supply.

The test stops at one window and one fund, with fills on minute bars and flat basis-point costs. It does not test a window in which the dollar fell for several years in a row, which would favour UDN holders and the rules that stay in.

Other currency 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.