Skip to content

Portfolio construction and performance

A source-checked guide to designing, implementing and evaluating portfolios through asset allocation, diversification, rebalancing, returns, benchmarks, risk and attribution.

Who this chapter is for — Readers who want to turn a set of instruments or strategies into a governable portfolio and evaluate it without confusing return, capital flows, risk, benchmark and skill. The path starts with the foundations and makes the conventions of professional reporting explicit.

Portfolio construction translates objectives, constraints and assumptions into implementable weights, instruments and rules. Performance measurement reconstructs what happened using coherent values, flows, costs, currencies, benchmarks and risk definitions. They are two halves of one process: a decision cannot be evaluated properly if it was never specified, and construction cannot improve if its outcome is not measured reproducibly.

This chapter organises established public knowledge in portfolio theory and investment performance. It does not prescribe a model portfolio, assign universal grades to metrics or introduce rules of the Emiciclo Method. The Emiciclo Method remains a separate, later layer that may build on these definitions without rewriting their sources or meaning.


Interactive map: from policy to attribution

Portfolio construction and performance measurement Twenty-four nodes from investment policy to explaining the result Portfolio construction and performance measurement Twenty-four nodes from investment policy to explaining the result Select a node to open its canonical article MANDATE AND CONSTRUCTION 1 Construction Objectives, constraintsand process 2 Asset allocation Policy and tactical tilts 3 Capital allocation Resources, liquidity andweights 4 Covariance Dependence in joint units 5 Correlation Standardised dependence DEPENDENCIES AND IMPLEMENTATION 6 Diversification Non-identical sources of risk 7 Concentration Shared and hiddenvulnerabilities 8 Efficient frontier Mean–variance trade-off 9 Rebalancing Weights, costs and discipline RETURNS AND CASH FLOWS 10 Portfolio return Prices, income, costs andcurrency 11 TWR Neutralises external cashflows 12 MWR Weights cash-flow sizeand timing 13 CAGR Equivalent compound rate 14 Benchmark Comparison set in advance ABSOLUTE RISK AND RETURN 15 Standard deviation Dispersion of returns 16 Maximum drawdown Peak-to-trough loss 17 Sharpe ratio Excess return pervolatility 18 Sortino ratio Return per downsidedeviation 19 Calmar ratio Return per maximumdrawdown COMPARISON AND EXPLANATION 20 Beta Estimated marketsensitivity 21 Jensen's alpha Intercept of a factormodel 22 Tracking error Dispersion of activereturns 23 Information ratio Mean active return pertracking error 24 Attribution Where the result camefrom Cyclepedia · educational diagram: state conventions, period and data
Explore both directions of the chapter: construct a coherent portfolio and measure its path under disclosed conventions.
Stage Decisive question Canonical entries
1. Define What purpose, horizon, acceptable loss, liquidity and benchmark govern the portfolio?
2. Structure How are capital, exposures and dependencies distributed?
3. Build Which weights remain compatible with estimates, constraints, cost and capacity?
4. Measure Which return describes the path, and how are external flows treated?
5. Compare Does the reference represent the mandate on a like-for-like basis?
6. Read risk and value added What risk sits in the denominator, and what relationship with the benchmark is being estimated?
7. Explain Which represented decisions produced the difference from the benchmark?

This is not a sophistication ranking. Every stage removes ambiguity from the next. A precise information ratio cannot rescue an unsuitable benchmark; an elegant optimiser cannot repair undocumented expected returns or covariances; a correct CAGR does not by itself distinguish investment return from deposits and withdrawals.


First direction: construct

1. Objectives and constraints before products

A portfolio exists to support a purpose: real growth, preservation, income, liability funding, absolute return or a result relative to a mandate. Purpose is translated into horizon, currency, liquidity needs, acceptable loss, leverage, tax, eligible instruments and decision rights before vehicles are selected.

Portfolio construction is therefore a cycle of policy, implementation, measurement and review. It is not synonymous with optimisation. A portfolio can be designed with simple rules and robust constraints; a numerically optimal solution can be impossible to implement because of lots, spreads, capacity, margin or tax.

Strategic asset allocation describes the long-term architecture. Where permitted, tactical allocation makes temporary deviations whose signal, size, horizon, risk and owner are defined in advance. Calling every observed deviation “tactical” after the outcome hides the difference between a decision and uncontrolled drift.

2. Weights, dependence and concentration

Capital allocation shows where resources are placed; it does not automatically show where risk resides. Nominal weight, volatility contribution, margin and scenario loss are different quantities. Their units, denominator and observation time must remain visible.

Covariance expresses joint movement in the units of the series; correlation standardises it. Both depend on the sample, frequency, window, currency and data treatment. They are estimated inputs, not permanent asset properties. Modest linear dependence in ordinary periods does not exclude common moves under stress.

Diversification is an effect of weights and dependence, not a count of tickers. Several positions may share an issuer, equity factor, currency, duration, liquidity source or counterparty. Concentration risk therefore needs several views and, where vehicles obscure the underlying exposure, look-through data.

3. Frontier and robustness

In mean–variance analysis, expected portfolio return depends on weights and expected returns; variance also depends on the covariance matrix. The efficient frontier contains portfolios that, under stated inputs and constraints, cannot increase expected return without increasing variance or reduce variance without sacrificing expected return.

“Efficient” does not mean suitable, stable or likely to win. Small input changes can produce large weight changes. Constraints, shrinkage, resampling, sensitivity analysis, Bayesian estimates and simpler construction rules address different aspects of estimation error; none eliminates it. Results should show input scenarios, turnover, cost and the gap between theoretical and executable weights.

Rebalancing returns exposures toward a policy that remains valid. It may be calendar-based, band-based or hybrid, but frequency and thresholds have no universal values. They depend on costs, volatility, liquidity, tax, cash flows and materiality. Rewriting policy is a different decision from rebalancing.


Second direction: measure

4. Return is not the ending balance

Portfolio return measures economic change within a perimeter. A reproducible value states opening and closing valuations, income, external flows, costs, taxes, base currency, FX, frequency and compounding. Price return, gross total return and net total return describe different perimeters.

When contributions or withdrawals occur, the method follows the question:

  • TWR geometrically links subperiod returns separated by flows and aims to reduce the effect of their timing on the measured strategy;
  • MWR finds the rate reconciling opening value, dated flows and terminal value and describes the experience of the capital actually invested;
  • CAGR turns an endpoint ratio into a geometric annual rate over a stated duration, but does not by itself handle interim flows or describe the path.

TWR and MWR can differ without either being wrong. Annualisation also depends on assumptions. Square-root-of-time volatility scaling and multiplication of period means are not identities for every series with autocorrelation, seasonality or overlapping returns.

5. The benchmark belongs to the specification

A performance benchmark is set before the result is judged. It should represent the mandate and universe and be measurable, unambiguous and comparable for currency, total return, costs, calendar and rebalancing. A famous index is not automatically the correct reference.

Policy benchmark, market index, hurdle rate, peer group and custom portfolio are distinct. The exact price/gross-total/net-total variant, local or base currency, hedge, data provider and methodology version must be stated. Changing reference after observing performance is a retrospective choice that changes the meaning of active return.

6. Metrics compress; they do not replace the path

Standard deviation measures dispersion of a chosen series. Maximum drawdown measures the largest observed fall from a peak to a later trough. One uses deviations around a mean; the other depends on return order. Neither is a maximum future loss.

Risk-adjusted metrics change their question when the denominator changes:

Metric Numerator Denominator What it does not establish by itself
mean differential return standard deviation of the same differential absence of tails, liquidity risk or skill
return above a target downside deviation relative to the target protection from every severe loss
stated annualised return observed maximum drawdown stability outside the selected window
mean active return tracking error causal alpha or an appropriate benchmark

No threshold makes one metric “good” across all markets, frequencies, horizons and processes. Ranking requires sample size, uncertainty, costs, leverage, liquidity, selection bias and the result distribution. If many strategies were tested and only the best is shown, even a high ratio may reflect data snooping.

7. Relationship to the benchmark and explanation of the result

Beta is the estimated slope between portfolio and benchmark returns. Jensen's alpha is the intercept of a specified model and sample—not a synonym for skill. Tracking error is the standard deviation of active return; information ratio relates the mean and dispersion of that same series.

All depend on benchmark, frequency, window, currency, risk-free treatment and model specification. An unsuitable benchmark can create apparent alpha; a linear beta need not capture convexity or tails; a low TE may be intentional or may come from a meaningless comparison.

Performance attribution then attempts to reconcile active return with decisions. Allocation–selection, factor, transaction and fixed-income models answer different questions. Arithmetic closure does not create causality: hierarchy, timing, currency, derivatives, cost and residual must remain visible.


Minimum contract for a comparable report

Coordinate Information to publish
Period opening and closing date/time, calendar, frequency and missing observations
Perimeter entities, accounts, assets, cash, collateral and liabilities included
Return simple or logarithmic; price or total return; compounding and annualisation
Flows external/internal classification, dates, values and TWR/MWR method or approximation
Economics gross/net basis, commission, spread, management, custody, tax and income
Currency base currency, FX sources, timing and hedges
Benchmark name, provider, variant, currency, rebalancing and rationale
Risk metric, formula, window, frequency, sample convention and ex-ante/ex-post status
Data sources, revisions, estimates, stale prices, corporate actions and controls
Governance owner, version, calculation date, exceptions and reconciliation

These coordinates also improve extraction by search engines and AI systems. A definition without unit, period or benchmark is easy to quote and easy to misuse. Cyclepedia prefers a conditional, verifiable answer to a simpler false one.


Errors this chapter prevents

  1. Selecting products before objectives and constraints.
  2. Calling a ticker count diversification.
  3. Treating expected returns and covariances as known facts.
  4. Publishing an optimal weight without costs, turnover and input sensitivity.
  5. Confusing deposits with return, or TWR with the capital experience.
  6. Comparing incompatible return variants, currencies or periods.
  7. Changing benchmark after seeing the performance.
  8. Inferring quality from a universal Sharpe, Sortino or Calmar band.
  9. Calling every outperformance alpha.
  10. Assigning causality to a decomposition that merely closes arithmetically.

Reading paths

If you are starting: portfolio construction → asset allocation → diversification → rebalancing → portfolio return → TWR and MWR → benchmark → standard deviation and drawdown → Sharpe ratio.

If you build or evaluate portfolios professionally: covariance → mean–variance and robustness → measurement contract → benchmark → TWR/MWR and annualisation → tracking error and information ratio → beta and alpha → performance attribution.

For aggregate, liquidity, counterparty and scenario risk, continue with Risk measurement and control. For budgets, contributions and sensitivities, use Risk allocation and sensitivities. For strategy and backtest validation, open Systematic trading, backtesting and validation.


Sources