The "Just One More" Paradox

A coin-toss game advertised as favorable produces an exponentially rising average wealth yet collapses the median and mode because wealth is multiplied, not added, each round.

The Setup

• 80% gain on heads, 50% loss on tails yields +15% arithmetic mean per flip (0:00-9:07).
• Starting at $100, two heads and one tail already leaves the player at $90 despite three favorable-looking bets (0:00-9:07).

Non-Ergodic Dynamics

• Population average grows exponentially while median and mode fall to $7.20 after 50 rounds across one million simulated players (0:00-9:07).
• The mismatch between ensemble average and time-average path defines a non-ergodic process (0:00-9:07).

Additive vs Multiplicative Betting

• Fixed-dollar bets produce an upward-sloping mode but waste edge when wealth changes (0:00-9:07).
• Betting a fixed fraction of current wealth lets the mode rise; the optimal fraction is given by the Kelly Criterion (0:00-9:07).

00:00:00The Just One More Paradox

00:00:00Overview

• The video opens with a deceptively simple coin-flip game that hides a counterintuitive mathematical principle about growth, averaging, and long-run wealth.
• The core tension is between arithmetic mean (easy to compute per flip) and the actual wealth trajectory individuals experience when outcomes compound multiplicatively over many trials.
• The narrative frames a paradox: each individual flip seems favorable, but the long-run outcome is poor for most people when wealth grows multiplicatively.
• The topic blends probability, statistics (mean, median, mode), logarithms, and practical investment strategy (Kelly Criterion).

"Today we’ll play a game that sounds simple, but reveals a shockingly counterintuitive mathematical principle."

00:01:00Setup of the Game

• You start with $100.
• Each coin flip is independent: 1/2 chance of heads, 1/2 chance of tails.
• Heads increases current wealth by 80% (multiplier of 1.8).
• Tails decreases current wealth by 50% (multiplier of 0.5).
• Example trajectory: 100 -> 180 (heads) -> 324 (heads) -> 162 (tails) -> ...
• The per-flip arithmetic mean of the multipliers is 0.5(1.8) + 0.5(0.5) = 1.15, i.e., a 15% average gain per flip if you averaged the outcomes linearly, not accounting for compounding.
• The naive takeaway might be: it’s a good bet, since the expected multiplier per flip is 1.15.

"There’s a ½ chance of 0.8 gain + a ½ chance of -0.5 loss. That’s an arithmetic mean of 0.15; a 15% average gain per coin toss."

00:01:40Non-Ergodicity: Population vs Individual Trajectories

• Simulation idea: a million people each start with $100 and play 50 rounds.
• The population average wealth grows exponentially (driven by rare, extreme luck on some trajectories). However, most individuals fare terribly.
• Key concept: non-ergodicity — the average across many independent trials (population average) does not reflect the typical outcome for a single agent (individual trajectory).
• The video states this as: the population average diverges from the median outcome of individuals.

"This is called a non-ergodic system, because the population average is different from the median outcome of individuals."

00:02:15Visualizing with Logarithms: Sequences and the Median/Mode

• The video uses a diagram: from the starting point, after 1 head and 1 tail, wealth drops from $100 to $90 (multipliers 1.8 then 0.5, or vice versa).
• In a tree of sequences, there are 20 possible sequences for 3 heads and 3 tails; the central point corresponds to equal heads and tails.
• The central sequence (e.g., 3 heads and 3 tails) is both the median (center of the distribution) and the mode (most common outcome) when looking at the logarithmic scale of wealth across all sequences.
• The logarithmic view reveals that many more sequences yield poorer outcomes than the best-case trajectories, but the long tail of very good outcomes pulls the arithmetic mean upward.

00:02:40The Paradox Emerges: The Just One More Paradox

• Exiting the logarithmic view, the mode (most probable wealth) slopes downward as flips accumulate, even though each flip appears favorable (1.8 gain on heads, 0.5 on tails).
• The paradox is that the more you flip, the worse the expected mode/median becomes, despite each flip being individually attractive.
• The explanation: multiplicative growth compounds the entire wealth; a small loss early reduces potential future gains dramatically.

"The mode slopes downward. This is shocking since every coin flip seems favorable." "This is called the “Just One More Paradox". Each individual flip seems good, but the outcome is bad."

00:03:40Why Multiplicative Growth Causes the Paradox

• The core mechanism is multiplicative: every flip multiplies current wealth by a factor (1.8 for heads, 0.5 for tails).
• As wealth decreases, the same percentage gain on subsequent flips yields smaller absolute gains, limiting how much you can recover.
• The paradox scales beyond gambling to life decisions: any scenario where outcomes compound multiplicatively can exhibit the same effect.

00:04:20A Different Strategy: Additive Betting as a Benchmark

• Instead of wagering your entire wealth on each flip (multiplicative strategy), try betting a fixed amount (e.g., $50) each time (additive strategy).
• In the additive approach: heads adds a fixed amount to wealth (e.g., +$40 if you bet $50 and win 0.8 of your bet), tails subtracts a fixed amount (e.g., -$25).
• Under this scheme, the growth rate curve changes: the mode now slopes upward and can align with the average, removing the extreme downward pull of multiplicative risk.
• However, the additive strategy is not perfect across different wealth levels (e.g., with less than $50 or much more than $50, scaling matters).

00:05:20Fractional Betting: Moving Beyond All-In and Fixed Bets

• The video proposes a more flexible idea: bet a fixed fraction F of your current wealth on each flip.
• On heads: wealth grows by a factor of (1 + 0.8F). On tails: wealth grows by a factor of (1 - 0.5F).
• If you evaluate sequences with a fixed fraction, you can derive the mode/median growth behavior under this strategy and compare to add/multiplicative extremes.
• Example intuition: with a small fraction, each flip’s impact is tempered, potentially preserving future upside.

00:06:05Deriving the Optimal Fraction: A Generalized Model

• Generalize with F as the fraction of wealth bet, B as the gain multiplier (1.8 for a head when betting a fraction F), A as the loss multiplier (0.5 for a tail when betting a fraction F), and P, Q as probabilities of heads and tails (each 0.5 in this video).
• The per-flip growth factor under a fractional strategy becomes a function of F, combining both outcomes.
• The transcript notes: the x-axis is the fraction F; the y-axis is the mode growth rate.
• Observations: when F = 0, no betting occurs; growth rate is 1 (no change). When F = 1, you bet your entire wealth and the growth rate is 0.949 (a loss on average per flip).
• The highlighted region shows positive growth; the problem is to find the F that maximizes the growth rate (the peak of the curve).

00:06:40Calculus to Find the Kelly Fraction

• The instruction states: differentiate with respect to F on the log scale (logarithmic differentiation) and set the derivative to zero to find the maximum growth rate with respect to F.
• The calculation yields the optimal fraction F* = 0.375 (37.5% of wealth).
• At this fraction, the mode growth rate per coin flip is 1.028 (i.e., a net positive growth per flip under the mode metric).

00:07:25Simulation Results with the Kelly Fraction

• A new simulation with a million people each betting F = 0.375 shows the average wealth trajectory and, importantly, the median increases steadily.
• This demonstrates that using the Kelly criterion (the optimal fractional betting rule to maximize long-run growth rate) improves the typical outcome (median) rather than just inflating the mean with a few lucky trajectories.
• The Kelly Criterion, formulated in 1956 by John L. Kelly Jr., is widely used to specify what fraction of wealth to bet to maximize long-term growth rate.

"The Kelly Criterion... maximizing growth rate." "Plugging in our values gives 0.375. That’s the fraction of our wealth we should bet to maximize growth rate."

00:08:10Historical Note and Practical Takeaways

• The Kelly criterion is an explicit, rigorous rule for fractionally betting wealth to optimize growth in repeated gambling/investment-like processes with favorable odds.
• The result (0.375) provides a practical target for the specific 1.8 vs 0.5 payoff setup and equal probabilities.
• The video emphasizes that the end result is not just a theoretical curiosity; it informs how to manage risk and growth in real-world decisions when outcomes compound.

00:08:40Credits and Related Resources

• The creator notes they wrote roughly 1100 lines of code to animate the video using the Manim graphics library (a derivative of the math animation work by 3Blue1Brown and the community).
• Acknowledges Jason Collins for his ergodicity economics blog post and points to Mutual Information as another explanation of the Kelly criterion.
• Links are provided in the video description (see Resources Mentioned).

00:09:00Closing

• The video invites viewers to reflect on probability, to recognize the paradox in multiplicative processes, and to appreciate how a math-driven approach (ergodicity, geometric/Kelly analysis) can reveal counterintuitive truths.
• Encourages further exploration of ergodicity economics and applied Kelly strategies.

Visuals and Diagrams (What to Recreate Visually)

• A wealth trajectory tree showing two initial branches: heads (1.8x) and tails (0.5x). Show how after several flips, some branches yield very high wealth while most decline.
• A log-scale visualization where the central region shows that equal numbers of heads and tails concentrate around a downward-trending mode, despite arithmetic mean growth.
• A linear scale comparison: multiplicative (full wealth on each flip) vs additive (fixed bet) vs fractional betting (bet F of wealth).
• A plot of growth rate vs fraction F, with a peak around F = 0.375, and annotation showing the corresponding per-coin-flip mode growth rate at the peak.
• A final panel showing simulated medians rising when using the Kelly fraction, demonstrating the practical benefit.

00:09:50Notable Takeaway Points (Emphasized ideas)

• The paradox arises from multiplicative compounding, not from the per-flip gain being negative.
• Arithmetic mean can be misleading in multiplicative processes; the geometric mean (log-scale growth) is a better guide to long-run growth.
• The Kelly criterion provides a principled way to balance growth and risk by selecting the optimal fraction of wealth to risk on each bet.
• A strategy that seems intuitively favorable on a single flip (betting a lot) can dramatically underperform in the long run due to early losses reducing future upside.

Resources Mentioned

• Code for the video: https://github.com/manforowicz/Manim-Videos/blob/main/kelly_criterion.py
• Jason Collins blog on ergodicity economics: https://www.jasoncollins.blog/posts/ergodicity-economics-a-primer
• Myers, J. K. (2021). Multiplicative Gains, Non-ergodic Utility, and the Just One More Paradox (Supplemental Information): https://www.researchsquare.com/article/rs-237495/latest.pdf
• Related video on the Kelly criterion: https://youtu.be/-9JM9suCIHs
• Manim library and 3Blue1Brown community (credited for visuals), Mutual Information YouTube channel

Final Thoughts

• The video provides an accessible, example-driven explanation of why a sequence of individually favorable bets can destroy long-run wealth when growth is multiplicative and time cumulates. The Kelly criterion serves as a practical remedy to maximize long-run growth while controlling downside risk, illustrating a fundamental intersection of probability, finance, and decision theory.