Backtest overfitting
Fitting a strategy to features of historical data so that past success transfers poorly to new observations.
Route · 26 · 20 terms
Portfolio
Portfolio risk, diversification and rebalancing.
Path goal: Match the task horizon to risk and define decision rules in advance.
Reading sequence
This editorial sequence preserves the Russian curriculum order and uses the existing translated guides.
Foundation
Core explanations: start here if the topic is new.
Portfolio risk: losses, horizons and co-movement
How to separate price fluctuation from unacceptable consequences and read educational stress scenarios without turning them into personal advice.
Diversification: look for independent causes, not just names
How to find hidden concentration and understand the limits of spreading risk through fictional examples.
Rebalancing and behaviour: restore the rule, not a market guess
How to distinguish restoring an educational structure from changing the objective, compare methods and account for costs.
Financial records: prepare the data without guessing the tax
Keep a transaction journal and supporting documents while separating financial arithmetic from jurisdiction-specific tax conclusions.
Further connections
Additional concepts and practical connections after the foundation.
Purchase price as an anchor: test the thesis instead of defending the old decision
Why an entry price can distort analysis through anchoring, loss aversion and confirmation seeking.
Backtest overfitting: how many attempts hide behind one beautiful curve
A beautiful historical curve may be the winner of thousands of random variants.
Budget and paper portfolio: a safe activity with an adult
Use tokens and cards without real trades, personal data, loss penalties or product recommendations.
Position sizing: measure the contribution of error, not confidence in the story
Position size translates the error of analysing one asset into consequences for the whole portfolio.
Currency prices, cash flows and country of risk
One fictitious company: revenue and debt currencies, quote
Expected value in decisions: probability and outcome size
Account for probability and outcome size, test decision robustness and avoid confusing positive expectation with acceptable risk.
Factor exposure: explaining returns is not promising a risk premium
A factor model describes joint return movements through selected risk sources or statistical characteristics.
Fat tails: why rare losses cannot be removed by an average
Why standard deviation does not describe all risk and how to build a rare-loss scenario without inventing a precise probability.
Forecast calibration: assess probability, not confident tone
How to record testable probabilistic forecasts, calculate the Brier score and distinguish calibration from informativeness.
Housing: the service of living, credit and comparison with renting
Model a mortgage, repairs, utility, concentration and liquidity without choosing housing for the reader.
Index and active approaches: compare implementation cost and purpose, not labels
An index approach seeks to reproduce a chosen benchmark; an active approach departs from it for its own purpose.
When IRR misleads: multiple roots, no solution and the wrong scale
Treat internal rate of return as an equation solution, not a universal measure of attractiveness.
Martingale and position increases: why a loss need not recover
How doubling after a loss differs from planned contributions, and why lowering average price does not remove capital risk.
Mean reversion and extrapolation: test the mechanism, not the shape of the series
Distinguish statistical regression to the mean, an economic recovery mechanism and an irreversible regime change.
Portfolio objectives and liabilities: build the spending calendar before choosing asset risk
How to evaluate a portfolio against future payments, dates, currencies, access and funding gaps rather than return alone.
The cost of rebalancing: turnover, spread and expenses matter more than elegant frequency
Rebalancing restores a rule, but every transaction has a price.
Verifying a financial source: date, scope and reproducibility
How to build a financial-evidence card, check dates and scope and explain a mismatch between two published figures.
Survivors and future knowledge: two errors that quietly improve a backtest
A historical test can be mathematically correct and still describe an impossible strategy.
Returns with contributions: money-weighted and time-weighted results
Why two correct percentages for one account can answer different questions.
Advanced topics
Specialised models and limitations: check assumptions against the foundation.
Correlation and covariance: why the result depends on the scale and window of the data
Correlation describes the linear joint movement of two series
Sequence of returns: why equal percentages leave different balances with withdrawals
Average return does not fully describe the outcome of a portfolio with regular withdrawals.
Volatility and drawdown: identical changes can produce different loss paths
Volatility measures return dispersion; drawdown measures the fall from a previous peak along a chosen path.
Key terms
Return correlation
A unit-free measure of the linear co-movement of returns over a defined set of observations.
Return covariance
An estimate of the joint deviation of returns that preserves the scale of their variability.
Diversification
Distributing investments across different sources of risk to reduce dependence on individual adverse events.
Return
The result of an investment relative to its value over a stated period and with a defined set of income and costs.
Financial risk
Uncertainty of an outcome, including loss or failure to reach a financial goal.
Investment horizon
The time until funds are needed for a specific financial goal.
Investment portfolio
A group of assets and positions considered as one whole for a defined financial goal.
Look-ahead bias
Using information in a historical decision before that information was published or available on the relevant date.
Maximum historical drawdown
The deepest decline from a preceding peak within a selected window and at a selected observation frequency.
Money-weighted return — MWR
A return measure that reflects the size and timing of the owner’s external contributions and withdrawals.
Portfolio turnover
A measure of transaction volume relative to portfolio value under an explicitly chosen formula.
Investor behavioral biases
Recurring decision patterns connected with emotions and the way information is perceived.
Asset allocation
Choosing the shares of asset classes and sources of risk in a portfolio in line with its objective and constraints.
Portfolio rebalancing
Returning a portfolio’s structure to preselected weights or ranges after they have changed.
Sequence risk of returns
The dependence of a portfolio result with cash flows on the order of market changes.
Survivorship bias
A distortion caused by studying only objects that remain instead of the original population.
Time-weighted return (TWR)
The linked result of subperiods between external cash flows.
Heavy tail
The part of a distribution where large deviations occur more often than the selected thin-tailed model assumes.
Volatility
The degree of variation in a price or return; the assessment depends on the period and measurement method.
Scenario checks
Real return
Compare nominal change with price changes.
Percentage change and CAGR
Calculate total change and annual growth.