Bitcoin price is expected to bottom at $35,000 in December, according to a model that measures the past two market highs.

Since it is clear that this cycle’s bull market is the highest, we created an updated halving model built on four Bitcoin cycles.

The model predicts a 72.5% drawdown from the cycle high of $126,219 to a cycle low near $35,000 in December 2026.

Inside the half-life framework

My last model correctly marked both the 2021 and 2025 top timeframes. New framework “akiba cycle model v2” combines a 50,000-run Monte Carlo simulation with walk-forward verification and leave-one-out cross-validation (LOOCV).

This cycle is divided into three linked components: the drawdown from a bull market high to the next cycle’s low, the number of days from a halving to that low, and the recovery multiple from a low to the next halving.

The drawdown and timing components had smaller historical errors than the recovery leg. This recovery leg caused the biggest miss in out-of-sample testing.

The model starts from the empirical pattern of previous cycles, where the peak-to-trough drawdown has softened with each epoch while still remaining deep.

Based on the cycle classification used in the accompanying chart, the historical drawdown from bullish high to cycle low has been 94.1% in the first cycle, 88.2% in the second cycle, 83.7% in the third cycle, and 77.6% in the fourth cycle.

The fitted forecast for the fifth cycle is centered around a 72.5% drawdown, and the simulated band is between 71.9 and 73.1%.

The drawdown distribution is tight because the monotonic decay is maintained across all four observations. LOOCV The root mean square error is 0.63 percentage points.

Using the bull market high of $126,219, the distribution of implied cycle lows is centered around the mid-$30,000s.

The median simulated low is approximately $34,700, with a range of $33,900 to $35,500 P10 to P90.

The timing points to the second half of 2026.

We also mapped out how long it takes for the market to reach a cycle low after a halving.

The number of days from half-life to cycle minimum increased stepwise from 778 days in cycle 1 to 784 days in cycle 2, then 890 days in cycle 3, and 923 days in cycle 4.

The forecast for the fifth cycle is centered around 980 days after the April 2024 halving, which corresponds to December 2026. The period from P10 to P90 is from November 2026 to January 2027.

The timing error of LOOCV is larger than drawdown, which is 37 days. This reflects differences in the extension pattern, including an increase of 6 days between the first two cycles.

Below is a summary view of the cycle history used in the model.

Multiple recoveries create the greatest uncertainty

The recovery leg is the part that the model considers to be the most unstable. This estimates the multiple from the cycle low to the next halving price, but this path has been compressed over time in historical series.

The multiples from H2 to the next half-life were 347.8x to H2, 67.2x to H2, and 20.8x to H2, with a median estimate of nearly 5.0x to H2.

Because this component has only three past observations and the walkforward test fails, the simulation uses a wide uncertainty band for the H5 halving price.

The range from 10 pesos to 90 pesos is $60,000 to $489,000, with a median price of $172,000.

I built and ran backtests myself, pressure-testing the model over previous cycles and revealing where its assumptions tracked reality and where they started to break down. Backtesting makes it clear where your approach worked.

When trained on cycles 1-3 and predicting cycle 4, the model produced a drawdown estimate of 78.2%. In contrast, the observed value was 77.6%, a difference of 0.7 percentage points.

Also, the minimum period of the cycle was predicted to be 929 days, compared to the observed 923 days, a difference of 6 days.

In terms of price, we predicted a cycle low of $15,012 versus the observed $15,474, resulting in a 3% error.

In the same exercise, the recovery multiplier was 38% lower (13.0x predicted, 20.8x actual). That mistake then propagated into a larger error in the implied halved price.

These diagnostics determine how the output is displayed.

The model treats the cycle low estimate as the main predictable variable and frames the next halving price as a scenario space.

The Monte Carlo engine samples from an ensemble of simple functional forms (linear fit, exponential decay, mean decrement variant), injects calibrated noise into the LOOCV residuals, and uses jackknife resampling of the four-cycle dataset to emphasize sensitivity to any one epoch.

It also clamps the output to the boundaries defined by the notes. The drawdown, timing, and recovery draw are then chained together to generate the joint allocation.

A snapshot of the distribution output for the fifth cycle is shown below.

The notes also include two probability statements derived from the set of simulated distributions. That is, under the model’s structural lower bound assumptions, there is a 64.4% probability that the H5 halving price is above $126,219 and a 100% probability that the cycle low price is above $20,000.

Both claims are conditional on model design, including small sample calibration and independence assumptions. This assumption treats drawdown, timing, and recovery as separable random draws, even though they can move together.

The observations underlying cycle classification help explain why the model focuses on drawdown and elapsed time rather than peak returns.

The peak gain relative to the previous halving price compressed in each era, rising from 10,375% in Cycle 2 to approximately 2,900% in Cycle 3 to 632% in Cycle 4.

Of note, the current cycle’s bullish high is set at 103% of the previous halving price.

At the same time, the half-life to peak interval was extended from 363 days after the first half-life to 522 days after the second half-life and 555 days after the third half-life.

Based on the selected data points, the model sets the fifth cycle bull market high 537 days after the April 2024 halving.

The model documentation lists several limitations that may change how these distributions are read.

Because it uses a total of four cycles, its tail may underestimate results outside the historical range.

It also does not take into account regime variables such as ETF flow patterns, custody structure, and macro-correlated inputs such as rates and liquidity.

The recovery module is flagged as the main source of uncertainty because the walkforward test showed that the cycle shape extrapolation did not capture the cycle 4 recovery fold.

For market participants who treat halving-era behavior as a reproducible template, the v2 framework formalizes two previous cycle regularities: drawdown rate drift and extension to cycle lows.

Leave the next halving price as a wide distribution rather than a point call.

The model’s median path sets the next cycle at a low level in the mid-$30,000s around December 2026. With the caveat that this is not financial advice, we are keeping the price of Halving 5 as an outcome band fixed at $172,000, in the middle of the $60,000 to $489,000 range.