Psychology of Bankroll Erosion and Chasing

Separate the Result from the Next Decision

Kahneman and Tversky’s 1979 prospect-theory paper studies choices under risk and how gains and losses are evaluated relative to a reference point. It is a foundation for thinking about loss framing, not evidence that every bettor will chase or that one intervention prevents it.

A prior loss does not improve the chance of an independent next outcome. Nor does an amount already lost make a new selection valuable. Shared information or correlated events can change a forecast, but those effects must be assessed rather than inferred from a desire to get back to even.

A model can also be wrong. Consistency is not a reason to preserve a forecast after credible contrary evidence appears, and a pause does not validate an assumed edge.

Price the Hypothetical Opportunity

Suppose a hypothetical selection pays +120, decimal 2.20, and a model assumes a 52% win probability. Break-even is 100/220 = 45.4545%. Expected net return is 0.52 × 1.20 − 0.48 = 0.144, or +14.40% of stake, under binary settlement without costs.

The 52% is an assumption, not a known “true” probability or an OwnTheLines model output. At an alternative 44% assumption, the same quote has 0.44 × 2.20 − 1 = −3.20% expected return. Increasing the stake multiplies exposure to both forecast error and outcome variation.

OwnTheLines does not identify market edges with a production prediction model. Its simulation limits can be used as practice constraints, but they do not establish that a personally chosen strategy is profitable.

A Fixed-Stake Drawdown Example

Assume an initial simulated bankroll of $10,000 and a fixed $100 stake on each of eight losing selections. The loss is 8 × $100 = $800, leaving $9,200, an 8% drawdown. Here $100 is 1% of the initial balance, not a continually recalculated percentage.

If instead the stake were exactly 1% of the remaining bankroll before each loss, with divisible amounts and no cent rounding, the balance would be $10,000 × 0.99^8 = $9,227.45, rounded to cents. The loss would be $772.55, or 7.7255%. These two policies should not be mixed in a ledger.

Doubling the next fixed stake to $200 risks another $200; even a win need not recover the $800 loss. There is no verified historical 5% ROI behind this scenario.

Pre-Commit a Review Rule

An illustrative policy might pause new selections for 48 hours after eight consecutive losses. Both numbers are chosen for this example, not validated optimal thresholds or a product-enforced feature. Keep the policy in an external note.

During the pause, inspect data errors, changed conditions, skipped rules, and the original forecast uncertainty. Continuing, revising, or stopping a simulation should follow that evidence. A break creates an opportunity to review; it cannot guarantee recovery or prevent all future losses.

Questions About Chasing and Tilt

Set the sizing assumptions in Bankroll Management 101, then review Psychology of the Bad Beat.

Q: What can signal chasing in a simulation?

A: Raising stakes or accepting selections outside the earlier rules primarily to recover losses. Compare the decision with the policy recorded before the result.

Q: Is the 52% probability in this example verified?

A: No. It is hypothetical. At +120 it implies +14.40% expected return before costs, but an assumed 44% instead gives −3.20%.

Q: Do eight losses at 1% always cost $800?

A: No. Eight fixed $100 losses cost $800. Recalculating 1% of a declining $10,000 balance before each loss leaves about $9,227.45 under divisible-stake assumptions.

Q: Must I always follow the model?

A: No. Investigate evidence of data errors, changed conditions, or poor calibration. Avoid changing a rule solely to chase a result, but do not mistake consistency for model correctness.