DeployQuantRSI(14) Mean Reversion on TLT
iShares 20+ Year Treasury Bond ETF: long-duration Treasuries, one of the most rate-sensitive bond ETFs. Backtest 2021-01-04 to 2026-10-02, $10,000 starting capital, computed by the same engine that runs live DeployQuant strategies.
The RSI(14) rule buys when the 14-day RSI is below 30 and sells when it is above 70. On TLT, the 20+ year Treasury fund, from 2021-01-04 to 2026-10-02, it turned $10,000 into $7,270. That is a CAGR of -5.40% and a max drawdown of 35.95%. Holding TLT did worse: -8.23% a year, a final value of $6,106 and a 42.77% drawdown.
The rule beat the fund by 2.83 points of CAGR and by 6.82 points of drawdown, and still lost almost 27.3% of the capital. It made 13 closed round trips with a 54% win rate, a profit factor of 0.47 and a position open at the end. It ranks 9th of 12 templates on TLT and 48th of 59 funds for this strategy.
The case is useful because it shows what a mean-reversion rule does when the asset keeps falling. TLT's price series lost 40.49% over the window. A fund that spends most of its sessions below its 200-day average does not give an oversold signal much to revert to, and the numbers below show where the rule lost and where it helped.
The test is one window, with daily-decision rules, no stop, no target and no fees in the headline run.
Year by year
| Year | RSI mean reversion | buy & hold |
|---|---|---|
| 2021 | −3.2% | −4.4% |
| 2022 | −26.5% | −30.2% |
| 2023 | 2.1% | 2.7% |
| 2024 | 1.7% | −7.7% |
| 2025 | 3.0% | 4.0% |
| 2026 | −4.4% | −7.7% |
Ahead of the fund in four years, behind in two
Held, TLT returned -4.4% in 2021, -30.2% in 2022, 2.7% in 2023, -7.7% in 2024, 4.0% in 2025 and -7.7% in 2026 to date. The rule returned -3.2%, -26.5%, 2.1%, 1.7%, 3.0% and -4.4%. It was ahead in 2021, 2022, 2024 and 2026 and behind in 2023 and 2025, by 0.6 and 1.0 points.
2022 is the year that decides the page. The fund lost 30.2% and the rule lost 26.5%, a gap in the rule's favour of 3.7 points that left it with a heavy loss anyway. One trade is responsible. The rule bought on 2022-01-04 at an adjusted 120.45, when RSI(14) was already under 30, and held until 2022-11-25 at 87.70. That is a 27.19% loss over 325 days, the worst trade in the table and the longest hold. The fund's own drawdown was just as steep, so the rule's early entry at the start of the year left it exposed for the whole decline.
2024 is the best relative year. The fund lost 7.7% and the rule made 1.7%, a gap of 9.4 points. Three trades closed that year and two won: 2.56% from 2024-01-19 to 2024-03-06, 5.23% from 2024-04-16 to 2024-05-16, and a loss of 4.51% from 2024-10-01 to 2024-12-05. The rule was in cash when the fund fell in the second half of the year, which is the reason for the gap.
2023 and 2025 were positive for both, and the rule trailed. In 2023 it made 2.1% on three closed trades: 8.40% from 2022-12-29 to 2023-01-19 (which closed in 2023), 4.30% from 2023-02-17 to 2023-03-14, and a loss of 9.86% from 2023-05-23 to 2023-11-09. In 2025 it made 3.0% against 4.0%. In 2026 to date it lost 4.4%, against -7.7% held, and the position opened on 2026-07-16 at 82.87 was 6.50% under water at the end of the data.
In five of six years the rule's return is within a few points of the fund's. The distinguishing years are 2022, when both lost heavily, and 2024, when the rule sidestepped a fall.
Month by month
| Year | Jan | Feb | Mar | Apr | May | Jun | Jul | Aug | Sep | Oct | Nov | Dec |
|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 2021 | 0.0% | −1.3% | −5.0% | 2.3% | −0.1% | 0.2% | 0.0% | 0.0% | 0.0% | 2.1% | −1.1% | 0.0% |
| 2022 | −0.8% | −1.6% | −5.2% | −9.2% | −2.1% | −1.3% | 2.4% | −4.4% | −7.9% | −5.9% | 6.8% | −0.2% |
| 2023 | 8.4% | 0.6% | 3.6% | 0.0% | 2.5% | 0.2% | −2.5% | −3.1% | −7.7% | −5.3% | 6.6% | 0.0% |
| 2024 | 3.1% | −2.2% | 1.6% | 0.2% | 4.9% | 0.0% | 0.0% | 0.0% | 0.0% | −6.2% | 1.9% | −1.2% |
| 2025 | 0.5% | 0.1% | 0.0% | 0.0% | 1.0% | 2.7% | −1.2% | 1.4% | 0.0% | 0.0% | 1.2% | −2.6% |
| 2026 | −0.0% | 3.3% | −0.5% | −0.8% | 0.5% | −0.3% | −1.9% | 0.7% | −5.2% | −0.1% | – | – |
The months that mattered
The month table shows zeros whenever the sleeve was in cash. The rule was invested on 57.1% of days, so most months show a number, and the largest ones are losses in the same months the fund fell.
The worst month was April 2022 at -9.21%, against -9.19% for holding. The rule was fully invested through it, so there is no difference. It was followed by September 2022 at -7.93%, September 2023 at -7.74%, October 2024 at -6.23% and September 2026 at -5.17%. Other large losses were October 2022 at -5.87%, March 2022 at -5.18% and October 2023 at -5.35%. September and October appear several times, and the table has only 5 or 6 observations per calendar month, so the point is only that the worst months came in autumn in this window.
The best months were January 2023 at 8.42%, November 2022 at 6.85% and November 2023 at 6.61%. The fund's best month was November 2023 at 9.49%. Each good month is a rebound after a run of bad ones. After a 7.93% loss in September 2022 and a 5.87% loss in October, the rule made 6.85% in November. After September 2023 at -7.74% and October at -5.35%, it made 6.61% in November. The rebound is short and the falls before it are longer, so the rule gives back more than it takes.
The run of four negative months from July to October 2023 shows how the rule behaves in a slow decline. It had been bought on 2023-05-23 at 87.13 and was held to 2023-11-09 at 78.54, a loss of 9.86%. The sell signal at RSI(14) above 70 is rare for this fund, which was above 70 on only 21 sessions in the whole window. In a persistent downtrend the rule buys on one oversold reading and has to wait for the other to exit.
Every trade
RSI mean reversion on TLT made 13 closed round trips and one position still open at the end of the test, an average hold of 86 days, an average winner of 3.90%, an average loser of −7.95%, a profit factor of 0.47, a longest losing streak of 2. It held a position at the close on 57.1% of trading days.
| Entry | Entry price | Exit | Exit price | Return | Days held |
|---|---|---|---|---|---|
| 2021-02-16 | $120.12 | 2021-06-03 | $115.11 | −4.2% | 107 |
| 2021-10-05 | $120.92 | 2021-11-01 | $122.06 | 0.9% | 27 |
| 2022-01-04 | $120.45 | 2022-11-25 | $87.70 | −27.2% | 325 |
| 2022-12-29 | $85.70 | 2023-01-19 | $92.90 | 8.4% | 21 |
| 2023-02-17 | $87.06 | 2023-03-14 | $90.80 | 4.3% | 25 |
| 2023-05-23 | $87.13 | 2023-11-09 | $78.54 | −9.9% | 170 |
| 2024-01-19 | $83.23 | 2024-03-06 | $85.36 | 2.6% | 47 |
| 2024-04-16 | $79.01 | 2024-05-16 | $83.14 | 5.2% | 30 |
| 2024-10-01 | $90.42 | 2024-12-05 | $86.34 | −4.5% | 65 |
| 2024-12-20 | $81.88 | 2025-02-04 | $81.25 | −0.8% | 46 |
| 2025-05-20 | $80.11 | 2025-08-05 | $83.33 | 4.0% | 77 |
| 2025-11-17 | $85.44 | 2026-02-18 | $87.03 | 1.9% | 93 |
| 2026-03-16 | $84.83 | 2026-06-10 | $83.82 | −1.2% | 86 |
| 2026-07-16 | $82.87 | open | – | −6.5% | – |
Prices are adjusted for splits and dividends, so they sit below the quotes printed at the time. An open position is marked at the last close.
The 14 trades
The rule made 13 closed round trips and opened a 14th on 2026-07-16. Seven closed with a gain and six with a loss. The average winner returned 3.90% and the average loser lost 7.95%, which gives a profit factor of 0.47. The median trade returned 0.94%, the average hold was 86.1 days and the median hold was 65.
A 54% win rate and a negative result come from the shape of the trades. The seven winners were small: 8.40%, 5.23%, 4.30%, 4.02%, 2.56%, 1.86% and 0.94%. The losers were fewer and larger: -27.19%, -9.86%, -4.51%, -4.17%, -1.19% and -0.77%. The rule has no stop that would have cut the 2022-01-04 trade short.
The best trade, 8.40%, ran from 2022-12-29 at an adjusted 85.70 to 2023-01-19 at 92.90, in 21 days, which is the shortest hold. It was sold at RSI(14) above 70. The second, 5.23%, ran from 2024-04-16 at 79.01 to 2024-05-16 at 83.14, and the third, 4.30%, from 2023-02-17 to 2023-03-14.
The first trade of the window, from 2021-02-16 at 120.12 to 2021-06-03 at 115.11, lost 4.17% over 107 days. A second one in 2021, from 2021-10-05 to 2021-11-01, made 0.94% in 27 days. Both prices, 120.12 and 120.45 in early 2021 and early 2022, are adjusted prices and are far above the 82.87 entry of July 2026.
Two trades on the loss side came from entering too early in a slide. The 2023-05-23 entry at 87.13 and the 2024-10-01 entry at 90.42 both lost, and both were bought at RSI(14) below 30 in a fund that kept falling. The last trade lost 1.19% from 2026-03-16 to 2026-06-10, and the open position from 2026-07-16 was 6.50% down.
By exit year the record was 1 win in 2 trades in 2021, 0 of 1 in 2022, 2 of 3 in 2023, 2 of 3 in 2024, 1 of 2 in 2025 and 1 of 2 in 2026.
Largest drawdowns
| Peak | Low point | Depth | Days to low | Recovered | Days to recover |
|---|---|---|---|---|---|
| 2021-02-17 | 2022-10-24 | −36.0% | 614 | not yet | – |
| 2021-02-12 | 2021-02-16 | −0.2% | 4 | 2021-02-17 | 1 |
Buy-and-hold's deepest drawdown ran from 2021-01-04 to 2023-10-19 and reached −42.8%.
One drawdown that never ended
The page lists a single meaningful drawdown: 35.95%, from a peak on 2021-02-17 to a low on 2022-10-24, which took 614 days to reach and had not recovered by 2026-10-02. The only other entry in the table is a 0.16% dip in February 2021. The rule never made a new high after February 2021.
Buy-and-hold's drawdown was 42.77%, from 2021-01-04 to a low on 2023-10-19, and it also had not recovered. The price series shows a deepest drawdown of 44.06% over the same dates, and a longest drawdown of 1443 sessions, with no recovery date.
The rule's drawdown is shallower than the fund's by 6.82 points, and the reason is the exit on 2022-11-25 and the cash periods in 2024. The low of the rule's curve came on 2022-10-24, a month before the sale of the 2022 position, which closed after the low and after the fund began to bounce. The fund's own low came later, on 2023-10-19, and the rule was invested through part of that second leg down with the 2023-05-23 trade.
The drawdown shows why a mean-reversion rule needs the asset to come back. On TLT it did not, and the rule recovered nothing. The equity curve ends at $7,270 with a 6.50% loss open. The description of this template says it plainly: in a persistent downtrend RSI can stay oversold for weeks while the position keeps losing, and there is no stop.
With trading costs
The headline run fills at the bar price. These runs charge slippage on every fill.
| Slippage per fill | CAGR | Max drawdown | Final value | Sharpe |
|---|---|---|---|---|
| None (headline) | −5.4% | −36.0% | $7,270 | -0.42 |
| 5 basis points | −5.6% | −36.1% | $7,170 | -0.44 |
| 10 basis points | −5.9% | −36.4% | $7,059 | -0.46 |
Costs are small next to the losses
With 27 fills the cost runs change the result by a small amount. At 5 basis points per fill the CAGR is -5.63%, the max drawdown is 36.15% and the final value is $7,170. At 10 basis points the CAGR is -5.89%, the drawdown is 36.36% and the final value is $7,059. The Sharpe ratio goes from -0.42 to -0.44 and then -0.46.
The cost of 10 basis points is about half a point of CAGR, which is small next to the 2.83 points by which the rule beat the fund. The rule still beats holding at both cost levels, since holding made -8.23% and the cost runs made -5.63% and -5.89%.
TLT is a liquid fund, with an average daily dollar volume of $2,159,787,473 and a median minute volume of 35,636 shares, so the cost assumption is probably conservative. The test applies one assumption to every fund and does not measure real execution. The cost is not what decides this result. The trade from 2022-01-04 does.
Changing the parameters
| Version | CAGR | Max drawdown | Round trips | Win rate | Final value |
|---|---|---|---|---|---|
| Published rules | −5.4% | −36.0% | 13 | 54% | $7,270 |
| RSI < 25 / > 70 | −4.9% | −34.0% | 11 | 55% | $7,480 |
| RSI < 35 / > 70 | −8.8% | −41.9% | 13 | 46% | $5,882 |
| RSI < 30 / > 65 | −3.3% | −26.0% | 17 | 59% | $8,269 |
| RSI < 30 / > 75 | −6.4% | −36.8% | 9 | 33% | $6,842 |
Four variants and what they say about the exit
The page tests four variants. Buying below 25 gives -4.93% with 11 trades and 6 wins, a max drawdown of 33.97% and a final value of $7,480. Buying below 35 gives -8.83% with 13 trades and 6 wins, a 41.88% drawdown and $5,882. A looser entry is worse here, since a fund in a downtrend gives more false signals as the threshold rises.
The sell threshold changes more. Selling above 65 gives -3.26% with 17 trades and 10 wins, a 26.00% drawdown and a final value of $8,269. That is the best variant and the only one close to a flat result. Selling above 75 gives -6.40% with 9 trades and only 3 wins and a 36.84% drawdown. A lower sell level takes small gains sooner, and TLT rarely rose far enough to reach 75.
The published 30/70 gives -5.40%. None of the four variants made money. The results run from -3.26% to -8.83%, which is a spread of several points on a fund where the rule has no edge. With 9 to 17 trades each, the spread is partly a matter of which trades happened to close before a decline. The sell-above-65 result should not be read as a recommendation, since it is a single outcome from one window. What holds across the variants is that every setting lost money on TLT.
How TLT behaved
| Measure | TLT |
|---|---|
| Data in this test | 2021-01-04 to 2026-10-02 (1444 sessions) |
| Total return, buy and hold | −40.5% |
| Annualized volatility | 15.3% |
| Deepest drawdown | −44.1% (2021-01-04 to 2023-10-19) |
| Up days | 49.0% |
| Average daily range | 0.94% |
| Average overnight gap | 0.54% |
| Correlation to SPY | 0.08 |
| Correlation to QQQ | 0.09 |
| Sessions above the 200-day average | 35.2% |
| Crossings of the 200-day average | 95 |
| Falls of 10% or more from a 20-day high | 3 |
Why TLT is a hard fund for this rule
TLT's price series fell 40.49% over 1444 sessions, with a CAGR of -8.64% and annualized volatility of 15.33%. Calendar returns were -4.53% in 2021, -31.21% in 2022, 2.79% in 2023, -8.07% in 2024, 4.23% in 2025 and -8.01% in 2026 to date. Only two of six years were positive and both were small.
The fund spent only 35.18% of sessions above its 200-day average and crossed it 95 times; 95 crossings in under six years is a choppy trend. Up days were 49.00% of sessions. The average up day was 0.74% and the average down day was -0.77%.
The RSI statistics are the core of the case. RSI(14) fell below 30 on 61 sessions. After those the median 5-day forward return was 0.02% and the median 20-day return was -1.63%, against baselines of -0.14% and -0.60% for all sessions. The 5-day figure is a little better than baseline, and the 20-day figure is worse. RSI(2) fell below 10 on 203 sessions with a median 5-day return of 0.03% and 20-day return of -0.65%. An oversold reading on TLT did not lead to a rebound over the next month, and the baseline itself is negative at -0.60% for 20 days. That is a description of a trending-down fund, and a mean-reversion rule needs the opposite.
TLT earned a negative overnight return, -70.11% on a log basis, and a positive intraday return of 18.78%. The overnight share of return was 136.58% and the intraday share was -36.58%. The fund lost its ground between the close and the next open. The rule buys at the open, so it was exposed to that overnight drift on every day it was held, and it was invested on 57.1% of days.
TLT's beta to SPY was 0.07 and its correlation to SPY was 0.08. It was perfectly correlated with TMF at 1.00, as TMF is its leveraged version, and 0.92 with BND and AGG, 0.91 with IEF and 0.86 with UST. The least correlated was TBF at -1.00, then RINF at -0.40 and KMLM at -0.39.
Among bond funds the rule did far better elsewhere. On IEI it made -0.20%, on AGG and BND -0.23%, on IGIB -0.48% and on IEF -0.84%. On SGOV it made 0.00% with no trades. TLT at -5.40% was the worst of the group and the median for the group was -0.23%. The shorter-duration funds fell less, and the rule lost less. The fund with the longest duration had the largest decline and the largest loss.
There were 3 falls of 10% or more from a 20-day high, over 5 days, so the fund's declines were slow rather than sharp, and the rule's oversold entries were not sharp-washout entries. The deepest single days were -3.47% on 2022-03-02 and 3.90% on the upside on 2022-11-10.
The rules
Buy when the 14-day RSI drops below 30 (oversold), sell when it recovers above 70 (overbought).
- WHEN the market opens · IF not invested AND RSI(14) < 30 · THEN buy with 98% of the sleeve
- WHEN the market opens · IF invested AND RSI(14) > 70 · THEN sell the whole position
A standard mean-reversion setup. The Relative Strength Index measures how stretched recent price action is. Readings under 30 have historically marked short-term washouts in uptrending assets. This template buys at the next session open and holds until RSI crosses back above 70. It has no profit target and no stop.
Good for: assets that trend up over time but overshoot on the way, such as broad index ETFs.
Watch out: in a persistent downtrend, RSI can stay oversold for weeks while the position keeps losing; there is no stop-loss in this template.
How the rules meet this fund
The rules are an entry at RSI(14) below 30 with 98% of the sleeve and an exit of the whole position at RSI(14) above 70, both at the open. There is no stop and no target.
On TLT the entry fires on slow, long declines and the exit rarely fires, because the fund was above 70 on 21 sessions only. The result is a long average hold of 86.1 days and a record that depends on whether a decline had finished when the rule bought. The rule bought into the 2022 decline on its first session of the year and held through it.
The other templates on the fund are listed on the TLT page. The RSI(2) snapback made 0.96%, the only positive result, and the EMA 12/26 trend made -0.67%. The dip buyer made -4.32% and the monthly cycle -8.08%. The RSI(14) rule sits at -5.40%, ninth of twelve. The strategy page lists all 59 funds, and the median across them is 2.98%.
The limits are the window, 13 closed trades and no fees in the headline run. A fund that rose over the period would have given the same rule a different result.
Build it from blocks (or type it in English), backtest it on 5.7 years of minute data in seconds, tweak any parameter, then paper trade it on live data. No card, no broker needed to start.
Frequently asked questions
Did RSI mean reversion beat buy-and-hold on TLT?
Over 2021-01-04 to 2026-10-02, RSI mean reversion on TLT returned −5.4% annualized vs −8.2% for buy-and-hold: it beat buy-and-hold by 2.8% per year, with a maximum drawdown 6.8 points shallower than holding (36.0% vs 42.8%).
How many trades did it make?
13 completed round trips over 5.7 years (27 fills), with 54% of round trips closing profitably.
Why RSI 30/70?
They are the conventional oversold and overbought bands from Welles Wilder's original formulation. Treat them as a starting point. In DeployQuant you can change them to 25/65 or anything else and re-backtest in seconds.
Does this strategy use a stop-loss?
No. The only exit is RSI recovering above 70. Adding a stop or a take-profit block is a one-block edit in the Lab.
Did RSI(14) mean reversion beat buy-and-hold on TLT?
Yes, but both lost money. The rule made -5.40% a year against -8.23% for holding, and had a 35.95% max drawdown against 42.77%. It ended at $7,270 against $6,106.
Why did the rule lose money with a 54% win rate?
The seven winners averaged 3.90% and the losers averaged -7.95%. One trade, from 2022-01-04 at an adjusted 120.45 to 2022-11-25 at 87.70, lost 27.19%. The profit factor was 0.47.
What was the worst trade on TLT?
The 2022 trade, held for 325 days with no stop, lost 27.19%. The next worst was -9.86% from 2023-05-23 to 2023-11-09. The rule was in the position for most of the 2022 decline.
Would a different RSI threshold have helped?
Selling above 65 gave -3.26% with 17 trades and a 26.00% drawdown, the best variant. Buying below 35 gave -8.83%. None of the four variants made money on TLT, so the loss is not an effect of the 30/70 setting.
How did the rule do on other bond funds?
Much better. It made -0.20% on IEI, -0.23% on AGG and BND, -0.48% on IGIB and -0.84% on IEF. TLT, the longest-duration fund, was the worst at -5.40%.
What did trading costs do on TLT?
The CAGR went from -5.40% to -5.63% at 5 basis points and -5.89% at 10, with 27 fills. The rule still beat holding at both levels. The loss came from the trades and not from costs.
Related
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.