After Costsresearch log
Crypto · On-chain flows

Stablecoin inflows and Bitcoin: a modest, recent effect

When total stablecoin supply jumps, Bitcoin tends to do slightly better over the following week. Since 2021 the effect is 1.6 to 2.3 percentage points and shows up in three data sources. Before 2021 it was a data artefact, and the simple rule did not beat holding Bitcoin.

30 September 2026 · 7 min read

Key findings

  • The signal: total stablecoin supply rises by more than on 90% of the previous year’s days. Buy Bitcoin the next day and hold for 7 days.
  • Since 2021, signal days were followed by 1.6 to 2.3 percentage points more Bitcoin return over 7 days than other days, in three versions of the data (t between 1.9 and 2.5). Adding past 1-, 7- and 30-day returns as controls barely changes it, so it is not dip buying in disguise.
  • Before 2021 the effect was a data artefact. The aggregator’s total jumped whenever it added a stablecoin to its coverage: 56 days with a “change” of more than ±3%, up to +47% in one day. With those jumps cleaned out, the 2018–2020 strategy fell from +40.2% a year to +0.9%.
  • The plain 7-day rule did not beat holding Bitcoin over its full history: 27.7% a year (Sharpe 0.83) against 43.5% (Sharpe 0.90) from May 2019, with a smaller drawdown (−62% against −77%). Its advantage is concentrated in 2023–2026.
  • There is no untouched hold-out for this signal. It now runs on an exchange demo account, and that is its first real out-of-sample test.

The idea

New stablecoins are minted when someone deposits dollars, and the most common reason to hold stablecoins is to buy crypto with them. A sharp rise in the total supply of stablecoins could therefore be “dry powder” that reaches Bitcoin within days. In the first round of research, a 30-day stablecoin-supply filter was also the only macro filter that helped in every period.

The rule

The first backtest looked good

The first run, on raw data at a lower assumed fee of 0.12% per side, had a Sharpe ratio between 0.90 and 1.06 in each of three periods, and in two of them about half of Bitcoin’s drawdown. A regression of the 7-day return on the signal, with past returns and the stablecoin trend as controls, gave +2.8% (t 2.3) in 2022–23 and +1.7% (t 2.1) in 2024–26 for Bitcoin, and similar numbers for Ether.

That was enough to build a bot around it. It was not enough to trust the data.

What the audit found

DefiLlama’s total is a sum over the stablecoins it tracks, and its coverage grew over time. Every time a coin was added, the total jumped without any money being minted. We found 56 days with a “daily change” of more than ±3%, all between 2018 and 2022, including +47.5% on 5 January 2019 and +36.6% on 3 April 2020. Compared with an independent series (CoinMetrics’ USDT plus USDC supply), the two sources disagreed by more than two percentage points on 105 days, and the correlation of their daily changes was −0.05 over the whole overlap and 0.42 since 2021.

We rebuilt the signal three ways: raw, cleaned (using the independent source on days where the two disagree), and from the independent source alone.

How much of the stablecoin effect survives data cleaning

Extra 7-day BTC return after a signal day compared with other days, by period and by stablecoin data source.

DefiLlama, rawDefiLlama, cleanedCoinMetrics USDT + USDC (independent)
Extra 7-day BTC return after a signal day (percentage points)
t-statistics in period order. DefiLlama raw: 1.12, 1.16, 2.18, 2.47. DefiLlama cleaned: 0.55, 0.63, 2.06, 2.24. CoinMetrics: −0.34, 0.60, 1.96, 1.93. Only the rows from 2021 on reach t ≈ 2. DefiLlama signals start in May 2019, so its first bar covers mid-2019 to 2020. Source: OLS coefficient of the signal-day dummy on the 7-day forward BTC return, HAC t with 7 lags.
Show data · values in %
DefiLlama, rawDefiLlama, cleanedCoinMetrics USDT + USDC (independent)
2018–20201.811.09-0.40
2021–20223.891.701.40
2023–20261.661.591.48
2021–2026 (since 2021)2.331.801.57
Data Period Signal days 7-day uplift t Uplift with controls t with controls
DefiLlama, raw 2018–2020 49 +1.81 pp 1.12 +1.65 pp 1.03
DefiLlama, cleaned 2018–2020 64 +1.09 pp 0.55 +0.95 pp 0.46
DefiLlama, raw since 2021 197 +2.33 pp 2.47 +2.13 pp 2.41
DefiLlama, cleaned since 2021 193 +1.80 pp 2.24 +1.59 pp 2.08
CoinMetrics USDT + USDC since 2021 174 +1.57 pp 1.93 +1.36 pp 1.78

The effect since 2021 survives cleaning and the independent source, at a smaller size. The pre-2021 evidence does not: the strategy’s 2018–2020 return fell from +40.2% a year (Sharpe 1.23) on raw data to +0.9% (Sharpe 0.26) on cleaned data. In the controlled regressions the coefficient on the past 7-day return was essentially zero (t between −0.93 and +0.29), so the signal is not a proxy for buying after a fall.

The full history against holding Bitcoin

Stablecoin-inflow signal vs holding Bitcoin

Buy BTC the day after a top-10 % daily rise in total stablecoin supply, hold 7 days; three versions of the input data against simply holding BTC.

raw DefiLlama data, 7-day holdthe bot's rule (±3% jump days ignored), 7-day holdcleaned data, 7-day holdthe bot's rule with a Parabolic SAR exit (exit chosen in-sample)Bitcoin buy and hold
Growth of 100 (log scale)
From 6 May 2019, the first day the 365-day threshold exists, to 28 Sep 2026, sampled every 14 days. Annual returns and Sharpe ratios: raw 30.6% (0.90), bot rule 27.7% (0.83), cleaned 12.6% (0.50), with the Parabolic SAR exit 47.8% (1.14), buy and hold 43.5% (0.90). Maximum drawdowns −62% to −66% for the strategies and −77% for buy and hold. Source: backtest with the bot's own signal code, Bitstamp daily closes, 0.25% fee per side.
Show data
Dateraw DefiLlama data, 7-day holdthe bot's rule (±3% jump days ignored), 7-day holdcleaned data, 7-day holdthe bot's rule with a Parabolic SAR exit (exit chosen in-sample)Bitcoin buy and hold
2019-05-06100100100100100
2019-05-20100100100100140.8
2019-06-03105.6100101.2100142.5
2019-06-17119.3119.3114.1119.3164.2
2019-07-01121.6105.298.45126.5186.0
2019-07-15118.1107.9101.0129.7190.8
2019-07-29116.798.0690.65117.9167.0
2019-08-12115.698.0690.65117.9200.2
2019-08-26110.098.0690.65117.9182.2
2019-09-09110.098.0690.39117.9181.3
2019-09-23107.398.0689.85117.9170.3
2019-10-07107.398.0689.85117.9144.3
2019-10-21107.398.0689.85117.9144.5
2019-11-04135.6120.1111.9148.9165.5
2019-11-18134.3120.1111.9137.6143.7
2019-12-02134.3120.1111.9137.6128.6
2019-12-16127.9120.1104.7137.6121.0
2019-12-30135.1120.1104.4137.6127.0
2020-01-13150.2120.1104.4137.6142.6
2020-01-27150.2120.1104.4137.6156.4
2020-02-10150.2120.1104.4137.6173.2
2020-02-24150.2120.1104.4137.6169.8
2020-03-09150.2120.191.16137.6139.6
2020-03-23150.2150.574.59172.4114.2
2020-04-06163.3147.481.12195.3129.3
2020-04-20148.1143.281.08176.7120.3
2020-05-04192.3156.892.27229.4156.2
2020-05-18213.6155.193.84251.1171
2020-06-01213.1145.991.63229.4179.5
2020-06-15213.1145.991.63229.4165.9
2020-06-29213.1145.991.63229.4161.6
2020-07-13213.1145.992.75229.4162.4
2020-07-27244.1167.2106.3262.8194.3
2020-08-10262.9180.1114.4283.0209.2
2020-08-24259.9178.1113.2279.8206.9
2020-09-07226.8155.498.75244.2182.5
2020-09-21227.6155.999.07245183.1
2020-10-05235.1161.0102.3253.1189.9
2020-10-19239.6164.2104.3268.7206.9
2020-11-02247.2169.4107.6309.8238.6
2020-11-16257.6176.5112.1330.5294.2
2020-11-30242.8166.3105.7349.7346.5
2020-12-14235.8161.5102.6358.2339.0
2020-12-28257.4176.3102.6502.5475.6
2021-01-11337.6231.3109.7658.9623.6
2021-01-25306.3209.899.51597.9567.4
2021-02-08468.6321.0152.2914.7816.4
2021-02-22547.0374.7177.71,068952.9
2021-03-08489.3362.4158.91,032922.1
2021-03-22504.6373.7163.91,065951.0
2021-04-05551.8408.7179.31,1641,040
2021-04-19490.8405.5155.01,117979.6
2021-05-03475.3359.1148.3989.01,006
2021-05-17359.4273.6113.0753.4766.8
2021-05-31325.4216.889.52645.4656.8
2021-06-14294.9216.889.52578.9712.9
2021-06-28294.9216.889.52578.9606.6
2021-07-12294.9216.889.52578.9582.3
2021-07-26294.9216.889.52578.9655.9
2021-08-09294.9216.889.52578.9814.2
2021-08-23294.9216.889.52578.9871.7
2021-09-06292.4215.088.77574.1927.1
2021-09-20292.4215.088.77574.1757.0
2021-10-04292.4215.088.77574.1866.1
2021-10-18292.4215.088.77574.11,091
2021-11-01292.4215.088.77574.11,072
2021-11-15285.8210.188.77561.11,119
2021-11-29270.6198.988.77531.21,017
2021-12-13270.6196.787.77525.3821.8
2021-12-27270.6194.186.59518.2892.1
2022-01-10238.1194.186.59518.2736.0
2022-01-24238.1195.986.59523.1646.0
2022-02-07238.1201.886.59539.0771.6
2022-02-21238.1201.886.59539.0652.1
2022-03-07238.1201.886.59539.0668.7
2022-03-21238.1201.886.59539.0721.9
2022-04-04238.1201.886.59539.0819.6
2022-04-18238.1201.886.59539.0717.9
2022-05-02238.1201.886.59539.0677.5
2022-05-16238.1201.886.59539.0524.7
2022-05-30230.7195.683.90560.7558.1
2022-06-13230.7195.682.52512.3394.9
2022-06-27230.7195.682.52512.3364.1
2022-07-11230.7195.682.52512.3350.9
2022-07-25230.7195.682.52512.3374.8
2022-08-08230.7195.682.52512.3419.0
2022-08-22230.7195.682.52512.3376.6
2022-09-05230.7195.682.52512.3348.1
2022-09-19230.7195.682.52512.3343.6
2022-10-03230.7195.682.52512.3345.4
2022-10-17230.7195.682.52512.3343.8
2022-10-31230.7195.682.52512.3360.5
2022-11-14205.7174.473.58456.8291.8
2022-11-28208.7176.974.66463.5285.1
2022-12-12208.7176.974.66463.5302.7
2022-12-26208.7176.970.63463.5297.6
2023-01-09208.7176.970.63463.5302.2
2023-01-23208.7176.970.63463.5403.2
2023-02-06208.7176.970.63463.5400.4
2023-02-20208.7176.970.63463.5437.0
2023-03-06208.7176.970.63463.5394.1
2023-03-20239.2202.878.25531.2489.2
2023-04-03244.2207.179.89531.2489.2
2023-04-17266.4225.987.16562.4518.0
2023-05-01280.8238.191.87557.4493.9
2023-05-15261.0221.385.39530.5477.8
2023-05-29261.0221.385.39530.5488.1
2023-06-12261.0221.385.39530.5455.6
2023-06-26299.3253.897.66608.3532.5
2023-07-10297.8252.597.15611.3535.1
2023-07-24293.3248.795.71602.2513.2
2023-08-07289.5245.594.46594.3513.3
2023-08-21278.2235.990.78571.1459.5
2023-09-04277.8235.690.64564.5454.1
2023-09-18276.4234.395.73561.5470.9
2023-10-02281.8238.997.36572.5483.8
2023-10-16274.4232.794.81557.5501.6
2023-10-30333.3282.6115.2677.2606.7
2023-11-13352.6298.9121.8716.3641.7
2023-11-27359.9305.1124.3731.1655
2023-12-11398.5337.9137.7809.7725.4
2023-12-25421.2357.1145.5855.8766.7
2024-01-08454.1385.0156.9922.7826.6
2024-01-22382.0323.9132.0776.1695.2
2024-02-05412.2349.5142.4837.5750.3
2024-02-19497.0421.4172.71,017910.9
2024-03-04656.0556.2227.91,3421,202
2024-03-18648.9550.2225.41,3271,189
2024-04-01668.9567.1232.31,3681,226
2024-04-15601.1509.6211.51,2301,116
2024-04-29604.8512.8212.81,2371,123
2024-05-13571.3484.4190.91,2131,107
2024-05-27592.0501.9197.81,3061,221
2024-06-10592.0501.9197.81,3061,222
2024-06-24592.0501.9197.81,3061,060
2024-07-08592.0501.9197.81,306997.5
2024-07-22594.1503.7198.61,3101,188
2024-08-05595.5504.9199.01,263950.0
2024-08-19572.8485.6191.41,2151,046
2024-09-02566.3480.1189.21,1681,040
2024-09-16570.2483.5190.61,1761,024
2024-09-30621.1526.6207.61,2791,113
2024-10-14618.3524.3206.71,2701,162
2024-10-28643.6545.7215.11,3221,230
2024-11-11812.6689.0271.61,6691,561
2024-11-25851.7722.1284.61,7491,636
2024-12-09892.2756.5298.21,8321,714
2024-12-23887.7752.6296.71,8231,667
2025-01-06922.9782.5308.41,8951,799
2025-01-20830.3704.0277.51,7051,797
2025-02-03849.3720.1283.81,7441,784
2025-02-17853.6723.8285.31,7531,685
2025-03-03853.6723.8285.31,7531,515
2025-03-17831.0704.6277.71,7061,478
2025-03-31787.2667.5263.11,6161,452
2025-04-14784.9665.5262.31,5301,488
2025-04-28793.9673.1265.31,5481,672
2025-05-12808.7685.7270.31,6741,808
2025-05-26808.7685.7270.31,7721,925
2025-06-09808.7685.7270.31,7721,940
2025-06-23808.7685.7270.31,7721,855
2025-07-07815.6691.6272.61,7821,905
2025-07-21812.6689271.61,9322,066
2025-08-04816.9692.62731,9422,024
2025-08-18800.1678.4267.41,9032,045
2025-09-01760.6644.9254.21,8091,921
2025-09-15800.2678.5267.41,9102,030
2025-09-29800.8679.0267.61,8622,012
2025-10-13743.8630.6248.61,7292,027
2025-10-27743.8630.6248.61,7292,007
2025-11-10743.8630.6248.61,7291,864
2025-11-24743.8630.6248.61,7291,553
2025-12-08743.8630.6248.61,7291,594
2025-12-22743.8630.6248.61,7291,557
2026-01-05743.8630.6248.61,7291,651
2026-01-19708.4600.6236.71,6471,628
2026-02-02682.3578.5228.01,5861,384
2026-02-16668.6566.9223.51,5551,211
2026-03-02708.4600.6236.81,5541,210
2026-03-16720.0610.4240.61,6941,317
2026-03-30720.0610.4240.61,5311,174
2026-04-13720.0610.4240.61,5311,309
2026-04-27748.8634.9250.31,5801,361
2026-05-11748.8634.9250.31,6691,438
2026-05-25748.8634.9250.31,6151,359
2026-06-08748.8634.9250.31,6151,109
2026-06-22748.8634.9250.31,6151,125
2026-07-06748.8634.9250.31,6151,126
2026-07-20724.8614.5242.21,6411,147
2026-08-03724.8614.5242.21,6091,116
2026-08-17735.2623.4245.71,6321,134
2026-08-31779.3660.8260.41,9891,382
2026-09-14751.4637.1251.11,8841,375
2026-09-28721.3611.6241.11,8091,459
Version, May 2019 – Sep 2026 Return a year Sharpe Max drawdown Time in market
Raw data, 7-day hold 30.6% 0.90 −66.1% 37%
The bot’s rule (±3% guard), 7-day hold 27.7% 0.83 −62.0% 37%
Cleaned data, 7-day hold 12.6% 0.50 −63.5% 39%
The bot’s rule with a Parabolic SAR exit 47.8% 1.14 −65.1% 46%
Bitcoin buy and hold 43.5% 0.90 −76.7% 100%

Over the whole history the fixed 7-day rule is at or below buy-and-hold on a risk-adjusted basis. Its weak spot was 2021: it made +2.6% while Bitcoin rose about 57%, with a −56% drawdown within the year. Its strong spot is 2023–2026, with a Sharpe of 1.20 to 1.28 against 1.15 and a drawdown of −21% to −34% against −53%.

Confirmations that did not confirm much

Exits

The rule sells after 7 days. We also tested about 40 alternatives: trend exits on daily and 4-hour candles, trailing stops, stop-losses and a vote of three trend indicators. The daily Parabolic SAR (0.02/0.2) had the best average Sharpe and the best worst-period Sharpe. Stop-losses never helped, trailing stops helped in the 2020–21 bull market and hurt in 2022–23, and the trend vote was rejected because its 200-day average made it too slow (a −43% drawdown in 2024–26 against −27%).

The Parabolic SAR exit was chosen on the same 2020–2026 data it is evaluated on, and its result depends on the candle source: its 2022–23 Sharpe is 1.22 on one exchange’s candles and 1.04 on another’s. Its better numbers in the table above are in-sample.

What runs live

Since 28 September 2026 the signal runs once a day in a scheduled cloud job and trades on an exchange demo account, 1,000 USDC per signal, long Bitcoin or cash. The main book uses the Parabolic SAR exit after the 7-day minimum. Shadow books track the fixed 7-day rule and a 20-day moving-average exit with the same entries, and a fourth book follows the direction model. The first entry was on 28 September 2026. Current positions are on the live results page.

Caveats