Does Floor-Kelly Help Savers and Retirees?

In history, full floor-Kelly added Bitcoin on top of monthly buying: a median 1.10× the Bitcoin of DCA into Bitcoin over 48 months. In simulated futures with the floor refit inside each path, that edge did not continue (0.90×). For a retiree starting at a cycle top, half floor-Kelly protected the portfolio, mainly by holding less Bitcoin.

Floor-Kelly With Money In and Money Out

Contributions, withdrawals and partner assets for a floor-anchored Kelly rule

Research on simulated worlds and walk-forward history. Not financial advice; no result here is a guarantee.

The Question

Paper 19 tested a Kelly rule whose growth input is the growth from today’s price to the power-law floor four years ahead. It tested a lump sum: money in once, nothing added, nothing taken out.

Real investors add money while they work and take it out when they stop. So this paper asks two things. Does the rule’s timing still add Bitcoin on top of monthly buying? And does it help a retiree survive a bad start?

It also compares the floor input with three others (the historical average return, the trend’s growth rate, and the trend itself as the target), at full, half and quarter Kelly, with gold, cash or US stocks next to Bitcoin.

The evidence comes from walk-forward history (2012 to 2026) and from 30-year simulated futures (16 cases, 2,000 paths each) in which price can stay below the floor. History has about three independent four-year windows and three cycle tops, so it describes rather than tests.


Saving: Money In

In history, full floor-Kelly added Bitcoin on top of DCA. Over every 48-month window from 2012, it ended with a median 1.10× the Bitcoin of putting each contribution into Bitcoin.

48 months, median vs DCA
1.10×
48-month windows ahead
78%
96 months, median vs DCA
1.27×

The gain is timing, not allocation. Against a constant mix at its own average weights, it gained 1.23× over 48 months and 1.76× over 96 months.

In simulation, with the floor refit inside each path, that edge does not continue. Full floor-Kelly ended 20 years of DCA at a median 0.90× in the base world, with outcomes widely spread (10th to 90th percentile 0.18× to 3.57×). It ended below 1 in four of five worlds.

20-year DCA, worldFloor-Kelly, refitFloor-Kelly, frozenTrend-anchored
BASE (cycles as fitted)0.90×3.34×4.40×
DAMPED (shocks halved)0.75×0.85×1.51×
FADING (shocks fade)0.45×1.28×1.97×
NO_REVERSION (random walk)0.15×1.00×1.00×
BROKEN (half the trend growth)3.01×4.34×4.85×

Median Bitcoin at the end of 20 years of DCA from today, against DCA into Bitcoin. Full Kelly, Bitcoin with gold, 2,000 paths per world. Refit: the floor is refit inside each path; frozen: held at its October 2026 fit.

Why: simulated crashes go deeper than history’s. In the base world 16% of months sit below the published floor, against 2% in 2012 to 2026. Each crash that sets a new low pulls the refit floor down, from 0.45× the trend today to about 0.24× after 25 years. Measured from that lower floor, the rule’s growth input is smaller, and it holds less Bitcoin.

Two panels from the base simulated world. Left: the refit 1% floor falls from 0.45 times the trend to about 0.24 times over 30 years, while the floor frozen at the October 2026 fit rises to about 0.57 times. Right: full floor-Kelly's average Bitcoin weight falls from 100% to about 16% with the refit floor and levels off near 48% with the frozen floor.
Inside the simulation, every crash pulls the refit floor down, and the rule holds less Bitcoin.

Trend-anchored Kelly led most simulated cases, but much of that lead is built in: its starting target equals the model’s own four-year forecast (0.898× trend). In history it trailed floor-Kelly at full Kelly (1.04× against 1.10×).


Retiring: Money Out

From 2012 to 2026 almost any start survived almost any withdrawal rate, because Bitcoin rose roughly 10,000-fold. The informative cases are retirements at a cycle top.

Retire at the topHighest rate survivedLowest real value at 8%
November 20139.7% / 19.1%11% / 44%
December 201712.3% / 20.0%19% / 56%
October 202112.6% / 20.8%19% / 51%

Each cell: Bitcoin only / half floor-Kelly, with gold and cash. $1M at the close of the top month, withdrawals indexed to inflation, to October 2026 (12.9, 8.8 and 5.0 years). These are survival rates over those windows, not 30-year rates.

Three panels, one per cycle top in 2013, 2017 and 2021. Real portfolio value at an 8% withdrawal for Bitcoin only and for full, half and quarter floor-Kelly. Bitcoin only falls furthest after each top; half floor-Kelly falls least.
Retiring at a cycle top, 8% a year: the gold sleeve paid the bills through the bear market.

The protection came mainly from holding less Bitcoin. At each top half floor-Kelly held only 23% to 29% Bitcoin. In simulation from 2.5× trend it beat Bitcoin only by 2.0 points of safe withdrawal rate with cycles intact and 3.5 points with trend growth halved, but by only 0.5 points with cycles fading. A constant mix at its own average weights came within half a point of it in all three worlds, and a fixed 50/50 Bitcoin and gold mix did as well or better in all three. The test that history and simulation agree here was adapted after the results were known.

Retiring today, near the floor, Bitcoin only did better. In the base world its 30-year safe withdrawal rate (95% of paths survive) was 19.5%, against 10.5% for half floor-Kelly.

These rates are high because the model assumes the power law holds (a median 480× rise in 30 years). With trend growth halved, Bitcoin only’s rate was 9.5%. In a random walk around the trend, Bitcoin only had no safe rate.


What a Reflecting Floor Assumed

Papers 15 to 17 estimated Bitcoin withdrawal rates in models that bounced price off a floor at 0.432× trend. Price can and does close below the floor. This study lets it stay there as long as the shocks take it, then runs the same shocks with the bounce added.

Bitcoin only, startNo reflectionReflecting
Starting today, near the floor19.5%30.0%
Starting at 2.5× trend5.0%6.5%

30-year safe withdrawal rate (95% of paths survive), base world, same shocks. Reflecting: price bounced off 0.432× trend.

A reflecting floor flatters Bitcoin retirement. It adds about 10 points from today, and about 30% more from a peak. The withdrawal figures in Papers 15 to 17 rest on that assumption.

A later study (Paper 23) corrected Paper 16 without it. In Paper 16’s own model, at its own start (the trend), a holder’s safe spending at 99% confidence fell from 103.6% to 46.1% of the floor’s growth. Its 80% “endowment” tier carried 12.3% ruin over 30 years instead of 0%. Paper 15’s figures, and any in Paper 17, have not yet been rerun.

History sits between the two models. Since 2012, price spent 2.3% of days below the published floor, and its longest stay was 66 days. The base simulation spends 16% of months below it; the reflecting model spends none.


What Holds, and What Does Not

Where history and simulation agree:

  • Savers did better with full floor-Kelly than with half or quarter floor-Kelly, at every risk aversion tested, in history and in four of five simulated worlds.
  • A retiree starting at a top did better with half floor-Kelly than with Bitcoin only, mainly because it held less Bitcoin.

Where they disagree: whether to aim Kelly at the floor or at the trend. The trend led in simulation, the floor in history. The paper makes no recommendation there.

“Normal” Kelly is close to Bitcoin only. Fed Bitcoin’s average return since 2010, it held 100% Bitcoin at full Kelly and 90% on average at half Kelly.

Stocks beat gold and cash as the partner, but only at 2012 to 2026 returns. With stock returns cut to 7% a year, cash led.

The floor zone was a 12-month buying signal in history. From below 1.25× the walk-forward floor, every 12-month return was positive (36 of 36). Over 48 months, against the trend’s own growth, it was a coin flip (16 of 32). This rests on three or four bear markets.

Pre-registered testVerdict
P1: full floor-Kelly beats DCA into BitcoinMixed: history passes, simulation fails
P2: half floor-Kelly helps a retiree starting todayNot supported
P3: floor-Kelly ranks first among Kelly inputsMixed: first at full Kelly in history, last in simulation
P4: the best partner assetStocks, sample-dependent
P5: the floor as a buying signalSupported

Read the Paper

36 pages, 16 figures, version 1.4. The method, the four Kelly inputs, saving and retiring in walk-forward history, 16 simulated cases, the partner assets, the reflecting-floor comparison, the scorecard against the pre-registered plan and every deviation from it.

The floor’s caution paid in history but not in a future of deeper crashes, and at a top it protected a retiree mostly by holding less Bitcoin.