Portfolio Management: Allocation, Rebalancing, Liquidity, and Position Sizing

Portfolio management turns investment ideas into positions that can survive uncertainty. Choosing a promising asset is only part of the task: how much to own, how to finance it, when to rebalance, and how to meet future cash needs can matter just as much.

These notes expand the supplied lecture outline into a practical mathematical lesson. Its numbered reference fragments lack a bibliography and are omitted. The research themes attributed to Dr. Jake Xia are discussed as supplied lecture claims, without inventing paper citations or treating unverified formulas as universal optimal rules. All numerical examples are hypothetical.

Learning goals

Translate investment objectives into allocations, constraints, and risk budgets.
Explain why institutional liquidity and spending needs shape portfolio choices.
Calculate mean–variance risk, cash combinations, and leverage effects.
Distinguish diversification from the separate decision to rebalance.
Recognize estimation error, asymmetric losses, and Sharpe-ratio limitations.
Derive binary Kelly sizing and identify when a gain–loss score is insufficient.

1. Portfolio construction is a sizing problem

Asset allocation chooses amounts across broad exposures. Strategy allocation sizes managers or investment approaches. Position sizing decides amounts in individual trades. Each is a version of the same question: how should limited capital and risk-bearing capacity be distributed?

Let V be portfolio wealth, pᵢ an asset’s price, and qᵢ its share holding. Its value weight is wᵢ=pᵢqᵢ/V. Include cash and financing consistently; a short position can have a negative weight. Notional, invested capital, and risk exposure are not interchangeable for derivatives or leveraged positions.

For one-period simple total returns r and fixed initial weights w, portfolio return is r_p=wᵀr when the full set of positions and financing are represented. With mean μ and covariance Σ:

E[r_p]=wᵀμ; Var(r_p)=wᵀΣw.

Sizing therefore shapes risk directly, but risk management is broader: it also covers liquidity, legal and operational controls, counterparty exposures, valuation, and contingency plans.

Objectives need a trade-off

“Maximize return and minimize risk” usually describes competing goals, not a single unconstrained solution. One common formulation is:

maximize wᵀμ−(λ/2)wᵀΣw, subject to feasible portfolio constraints.

The parameter λ>0 determines the return–variance trade-off. Another formulation minimizes variance subject to a target expected return. Different loss functions, liabilities, horizons, and preferences imply different objectives.

Constraint or budget	What it controls
Position and issuer limits	Concentration in individual names or linked exposures
Sector, country, currency, and factor limits	Hidden common risks across differently named assets
Gross leverage, borrowing, and collateral limits	Financing dependence and potential margin calls
Liquidity reserves and redemption limits	Ability to meet cash needs without forced sales
Spending or liability schedules	Timing and currency of required payments
ESG or other mandate exclusions	Eligibility under the institution’s investment policy
Turnover and tax budgets	Implementation drag and constraints on change

Constraints are not necessarily arbitrary. Many reflect real obligations, model uncertainty, or governance. Their costs and rationale should be explicit rather than hidden in an optimizer.

2. The institutional endowment problem

An endowment commonly seeks to support current spending while preserving purchasing power for future beneficiaries. The investment portfolio is part of a larger institution whose revenues, expenses, donations, and borrowing capacity also change during stress. Yale’s own endowment overview frames investment management together with sustainable spending.

Strategic choices

Choice	Potential benefit	Practical challenge
Internal versus external management	Control and specialization versus access to outside expertise	Staffing, fees, oversight, and capacity
Public versus private markets	Liquid price discovery versus access to different investment opportunities	Lockups, capital calls, valuation uncertainty, and exit timing
Passive beta versus active management	Low-cost benchmark exposure versus potential excess returns	Manager skill assessment, fee drag, and benchmark choice
Direct investment versus funds of funds	Direct control versus broader manager access	Due diligence resources and additional fee layers

Bottom-up manager selection can be important, especially where access and skill vary. It does not replace top-down exposure, liquidity, and liability management. Several managers with different labels can still share equity beta, leverage, or dependence on the same exit market.

Private assets’ infrequent marks can smooth reported volatility without making their economic risks disappear. Unfunded commitments are contingent liquidity demands, not merely future allocation opportunities.

A liquidity stress example

Suppose an institution has a 100 million portfolio: 20 million in cash and high-quality liquid assets, 30 million in public equities, and 50 million in private assets. Over the next year it may need 5 million for spending, 8 million for private-fund calls, and 3 million for collateral or other cash requirements.

That is 16 million of cash needs, leaving a 4 million buffer against the initial liquid reserve. If public assets fall, private distributions stop, or calls arrive sooner than expected, the institution cannot assume that selling private holdings at their last reported value is feasible.

Cash-flow matching asks whether usable resources arrive in the right amount, currency, and time. An infinite institutional horizon does not eliminate next month’s cash deadline. The lecture cites the 2008 crisis as a reminder of this issue; a specific historical performance claim would require the institution’s actual records.

3. Diversification and mean–variance theory

For two risky assets with weights w and 1−w:

σ_p²=w²σ₁²+(1−w)²σ₂²+2w(1−w)ρσ₁σ₂.

If both have 20% volatility, an equal-weight portfolio has volatility 14.14% at zero correlation and 20% at perfect positive correlation. Imperfect co-movement reduces combined variance relative to the perfectly correlated case. The benefit comes from joint behavior, not the number of line items alone.

The efficient frontier describes portfolios with the highest expected return for a given variance, or lowest variance for a specified expected return, within the feasible set and chosen inputs. It is an ex ante model, not a promise about realized outcomes.

Diversification can reduce some risks immediately when holdings are combined. It does not require systematic rebalancing to exist, and it does not eliminate market-wide losses. Rebalancing is a separate policy that maintains a desired exposure or changes it according to a stated rule. Investor.gov’s allocation guide treats allocation, diversification, and rebalancing as related decisions.

4. Cash, the capital allocation line, and leverage

Suppose a risky portfolio has expected return μ_P and volatility σ_P, and a risk-free asset offers r_f over the same horizon. Allocate a fraction a to the risky portfolio and 1−a to cash:

E[r]=r_f+a(μ_P−r_f),

σ=abs(a)σ_P.

The absolute value matters if a is negative. For a≥0, this gives the capital allocation line:

E[r]=r_f+σ(μ_P−r_f)/σ_P.

Its slope is the ex ante Sharpe ratio of that risky portfolio under the stated risk-free convention. Any risky portfolio gives a CAL; the tangency portfolio provides the highest such slope in the idealized feasible mean–variance setting. The capital market line adds equilibrium assumptions and is not merely another label for every CAL.

Worked allocation and borrowing

Take r_f=3%, μ_P=9%, and σ_P=12%:

Risky allocation a	Cash / borrowing weight	Expected return	Volatility
0.5	+0.5	6%	6%
1	0	9%	12%
2	−1	15%	24%

For a=2, the idealized expected return is 2μ_P−r_f=15%. This assumes borrowing at the same risk-free rate, no fees, stable exposure, and no binding margin constraints. If borrowing instead costs 5%, the expected return becomes 2×9%−5%=13%, before other costs.

Risk-free means certainty for a specified payoff and horizon under the model. Rolling short-term cash can have reinvestment risk, and a nominally certain payoff can still face inflation or currency risk relative to an institution’s liabilities. Leverage adds path-dependent funding and liquidation risk even when a one-period variance formula looks simple.

5. Rebalancing: correct the two-year example

Start with 100, split equally between A and B. Asset A returns +100% then −50%; B returns −50% then +100%. Ignore costs, taxes, and distributions.

Buy and hold

Date	A value	B value	Total
Start	50	50	100
After Year 1	100	25	125
After Year 2, no rebalance	50	50	100

Each asset ends where it started, so the buy-and-hold portfolio’s two-year cumulative return is 0%.

Restore 50/50 after Year 1

After the first year, sell 37.5 of A and buy 37.5 of B. Each position is now 62.5. In Year 2, A becomes 31.25 and B becomes 125, giving total wealth 156.25.

The equal-weight portfolio earns 25% in each year, so its two-year cumulative return is 1.25²−1=56.25%. The supplied outline’s 25% is the one-year portfolio return, not the two-year compounded return.

This particular alternating path rewards selling the relative winner and adding to the relative loser. It is not an arbitrage or a universal rebalancing premium. In a sustained trend, rebalancing can sell the asset that keeps outperforming. Trading costs and taxes can consume gains.

Choosing a policy

Calendar rebalancing trades at fixed intervals. Threshold rebalancing acts when weights move beyond tolerance bands. Cash-flow rebalancing directs contributions or withdrawals toward restoring the target before trading existing positions. Compare these policies on risk control, turnover, taxes, and liquidity rather than assuming the highest frequency is best.

6. Practical limitations of mean–variance optimization

Estimation error and unstable allocations

Expected returns are difficult to estimate precisely, and correlations can shift. In a simple unconstrained excess-return problem, an optimizer produces a risky exposure proportional to Σ⁻¹(μ−r_f1). Weak covariance directions and small expected-return changes can therefore create large allocation changes.

Useful responses include shrinkage, robust or scenario-based inputs, turnover penalties, diversified reference portfolios, and documented concentration or leverage limits. Constraints can reduce sensitivity but may also hide a bad input model. Evaluate the stability of the final portfolio, not merely the numerical optimality of one fitted solution.

Volatility is one risk measure, not the whole objective

Variance treats deviations above and below the mean symmetrically. Investors may instead care about shortfall against spending needs, permanent capital loss, drawdown, tail loss, or forced-liquidation risk. A distribution with occasional large losses can have an attractive historical standard deviation until the loss occurs.

Options illustrate why risk must be defined carefully. For an otherwise fixed vanilla option in a standard pricing framework, higher implied volatility generally raises its value, including for a long out-of-the-money call. But the premium can still be lost completely; time decay, entry price, realized volatility, and the underlying path matter. Higher portfolio return volatility is not automatically beneficial just because a holding has positive vega.

Sharpe ratio does not contain every preference

The Sharpe ratio (μ−r_f)/σ ignores skewness and other distributional details, and estimated values depend on horizon, dependence, and sample quality. Selling crash protection can look attractive by this metric before a rare severe loss.

Under idealized equal-rate financing, scaling a risky portfolio preserves its Sharpe ratio. Thus Sharpe alone does not determine how much risk an investor should take. However, it is too strong to say mean–variance theory has no sizing formula: the objective with risk aversion λ explicitly chooses size. The issue is whether that objective and its inputs represent the investor’s actual needs.

7. Gain–loss scores versus optimal position size

The supplied lecture attributes the score

s=(G−L)/(G+L)=1−2L/(G+L)

to Dr. Jake Xia’s gain–loss sizing discussion. For nonnegative G,L with G+L>0, it lies in [−1,1]. If both are zero, it is undefined. Before treating it as an investment size, define exactly what G and L measure and what objective is optimized.

One possible definition for an excess return R is G=E[max(R,0)] and L=E[max(−R,0)]. Then:

G−L=E[R], and G+L=E[abs(R)].

The score is therefore expected return divided by expected absolute return under this definition. It measures normalized signed advantage, but it does not generally solve expected-log-wealth maximization or a drawdown-constrained investment problem.

Deriving Kelly for a binary payoff

Assume a fraction f of wealth is exposed to a payoff of +b with probability p and −a with probability q=1−p, measured per unit of exposed capital. Remaining wealth earns zero in the chosen numeraire. Terminal wealth multipliers are 1+fb and 1−fa, which must remain positive for log utility.

Kelly maximizes long-run expected log growth under the specified repeated-bet assumptions:

g(f)=p log(1+fb)+q log(1−fa).

At an interior optimum:

g′(f)=pb/(1+fb)−qa/(1−fa)=0,

giving *f=p/a−q/b=(pb−qa)/(ab)**.

Apply actual borrowing, shorting, and exposure constraints afterward as part of the feasible optimization; a prohibited negative or leveraged position is not an implementable solution. This derivation assumes known probabilities and payoffs, a specified wealth objective, and feasible rebalancing between repeated bets.

When the lecture score matches Kelly—and when it does not

For an even-money bet with a=b=1, G=p and L=q, so the lecture score is p−q=2p−1, exactly the unconstrained Kelly fraction in that setup. With unequal payoff magnitudes, the general formulas differ.

Example: win 20% or lose 10%, each with probability one half. Then G=0.10, L=0.05, so s=1/3≈33.33%. But Kelly gives f=0.5/0.1−0.5/0.2=2.5*, or 250% exposure, if such leverage is allowed. Under a long-only 0≤f≤1 constraint, the constrained Kelly optimum is f=1, not one third.

Multiplying every payoff by a positive scale c leaves s unchanged, while general Kelly exposure scales as 1/c. That alone shows why the score cannot be a universal optimal capital fraction. If “gain” and “loss” instead mean unweighted conditional payoff averages, the score may omit win/loss probabilities entirely.

The classical Kelly exposition and risk-constrained Kelly research make the objective and constraints explicit. Fractional Kelly scales a fitted Kelly exposure down, often to mitigate uncertainty or drawdown sensitivity, but it is not a guarantee against ruin from model error or omitted risks.

8. Sizing by loss budgets and scenario constraints

A practical size can start from the loss an institution can bear. If a position is assumed to lose 8% of its notional in a stress scenario, and the portfolio’s permitted loss contribution is 1% of wealth, a simple standalone cap is:

position notional / wealth ≤0.01/0.08=12.5%.

This is a scenario-based cap, not an optimum or a promise that actual loss cannot exceed 8%. Include overlapping scenario losses across holdings, financing, gaps, and liquidity. A stop order does not guarantee its trigger price during a market gap.

For volatility budgets, a position’s marginal variance effect depends on Σw, not its own volatility alone. Different ways to allocate risk—equal capital, equal marginal risk contribution, factor budgets, or tail-loss budgets—answer different questions. They need not select the same portfolio.

The appropriate process connects estimated opportunity, plausible adverse outcomes, constraints, and the investor’s utility or spending requirements. No single ratio replaces that full decision.

9. Crowd feedback, bubbles, and power laws

The supplied outline describes two additional research themes attributed to Dr. Xia: self-organized crowd behavior and unequal capital growth producing dominant market agents. Without the underlying papers, the discussion here is conceptual rather than a claim to reproduce a particular theorem or empirical result.

Feedback can make prices and risk endogenous

Trend-following demand can reinforce a price increase; performance attracts capital, which funds more buying. The reverse can occur when losses trigger margin calls, risk cuts, redemptions, and sales. Fundamentals can still matter—the point is that trading and financing feedback can amplify or alter the adjustment path.

A strategy can appear liquid when few investors use it and become crowded when many hold similar positions. Portfolio risk models based only on historical price correlations may miss correlated future liquidation decisions. Stress tests should therefore consider who else holds the exposure and what financing rules could force them to sell.

Compounding does not automatically prove a power law

A stylized capital process might be Aᵢ,ₜ₊₁=Aᵢ,ₜ(1+fᵢ,ₜRᵢ,ₜ₊₁), subject to survival, flows, and constraints. Heterogeneous opportunities and compounding can create unequal wealth outcomes, but they do not automatically imply a power-law tail or a literal winner-take-all equilibrium.

For example, simple multiplicative growth can yield approximately lognormal outcomes under suitable log-growth assumptions. Power-law tails require additional structural conditions in particular models, such as certain entry, reinjection, or multiplicative/additive dynamics. A model-generated tail law is not proof that every market’s wealth distribution follows it.

The portfolio lesson is to examine crowding, capacity, survival bias, and feedback alongside static expected returns—not to assume that joining a popular strategy ensures compounding success.

10. A practical portfolio-management sequence

Define the horizon, spending or liability needs, risk capacity, loss tolerance, and permitted investments.
Map all holdings and managers into shared economic exposures, currencies, liquidity terms, and financing obligations.
Build a liquid reserve plan and stress capital calls, margin, withdrawals, and reduced distributions together.
Estimate returns and risks with uncertainty ranges; compare optimized weights with simple diversified references.
Choose an explicit sizing objective and limits. Validate any gain–loss or Kelly rule against its payoff definitions and leverage assumptions.
Set a rebalancing policy with tolerance bands, trading costs, taxes, and cash-flow use specified in advance.
Evaluate performance against appropriate benchmarks and risk budgets, separating market exposure, active choices, and implementation costs.
Review after new information or mandate changes; avoid rewriting the strategy merely to explain the most recent winning asset.

11. Practice questions

Two uncorrelated assets each have volatility 20%. What is the volatility of an equal-weight portfolio?
With cash at 3% and a risky portfolio at expected return 9% and volatility 12%, what are expected return and volatility at a=0.5?
If a=2 but borrowing costs 5%, what is expected return before other costs?
In the alternating +100%/−50% two-asset example, what is the two-year cumulative return with annual restoration to 50/50?
Does a buy-and-hold diversified portfolio have no diversification benefit merely because it is not rebalanced?
For an even-money binary bet with p=0.55, what is the unconstrained Kelly fraction? How does this differ from the unequal-payoff example above?
Why can an institution with a perpetual horizon still need a large liquid reserve?
Does simple multiplicative compounding guarantee a power-law wealth distribution?

Answers

√[0.5²×0.2²+0.5²×0.2²]≈14.14%.
6% expected return and 6% volatility under the stated single-period model.
2×9%−5%=13%.
56.25%, from 1.25²−1, before costs and taxes.
No. Diversification comes from combining exposures; rebalancing separately controls how those exposures evolve.
2p−1=10%. With unequal gains/losses, payoff magnitudes enter the Kelly formula and the normalized gain–loss score generally does not equal it.
Spending, capital calls, collateral, and withdrawals have near-term deadlines even if the long-run mission continues indefinitely.
No. The tail distribution depends on the model’s additional assumptions and mechanisms.