Video summary

Lecture 2 contd. - Part 2

Main summary

Key takeaways

Educational

Main ideas and concepts

  • Modeling multiple stocks with random variables

    • For n stocks, each stock’s return is treated as a random variable.
    • You have:
      • n expected values (means) — “first moments”
      • n variances
      • n × n covariance/correlation relationships summarized by a matrix.

  • Covariance and correlation relationships

    • Covariance between two random variables (returns of two stocks) is defined using deviations from their expected values:
      • (\text{Cov}(X,Y)) uses terms like ((r_1 - E[r_1])) and ((r_2 - E[r_2]))
      • then takes an expected value of the product of those deviations.
    • A covariance matrix:
      • Diagonal elements = covariance of each stock with itself = variances
      • Off-diagonal elements = covariances between different stocks
      • Symmetry: (\text{Cov}(i,j) = \text{Cov}(j,i)) (described as a “mirror image”).
    • Correlation coefficient
      • Correlation between two variables equals covariance divided by the product of standard deviations:
        • Numerator: covariance
        • Denominator: ( \sigma_X \sigma_Y )
      • It’s described as a measure of the “relationship” between two stocks/variables.

  • Copulas (mentioned as a future topic)

    • Copulas are introduced briefly as linkage functions that describe dependence between random variables beyond simple correlation.

  • Portfolio construction: weights, expected return, and risk

    • A portfolio is formed by allocating capital across assets using weights (w_i).
    • If total wealth is (W), then (w_i) represents the fraction of wealth invested in stock (i).
      • Example: investing 10 out of 100 → (w = 0.1) (10%).
    • Constraints discussed:
      • Typically weights sum to 1 when short selling is not allowed.
      • With short selling, some weights may be negative, but the sum can still be constrained to 1 (via normalization discussed later).

  • Expected return of a portfolio

    • Portfolio expected return:

      • [ E[R_p] = \sum_{i} w_i \, \bar{r}_i ]

    • (\bar{r}_i) is a sample average when population expected values are not known.

  • Portfolio variance (risk) using the covariance matrix

    • Portfolio variance is computed using a double summation over weights and covariances:

      • [ \sigma_p^2 = \sum_{i=1}^{n}\sum_{j=1}^{n} w_i w_j \,\text{Cov}(i,j) ]

    • Interpretation:

      • Diagonal terms contribute (w_i^2 \sigma_i^2)
      • Off-diagonal terms contribute (w_i w_j \text{Cov}(i,j))
    • Connection to correlation:
      • [ \text{Cov}(i,j) = \rho(i,j)\,\sigma_i \sigma_j ]

  • Risk–return visualization and the portfolio optimization goal

    • Portfolios are conceptualized on a Cartesian graph:
      • x-axis: return (first moment)
      • y-axis: risk (variance/standard deviation, second moment)
    • Goal: find weights that match an investor’s objective:
      • maximize expected return
      • minimize variance/risk
      • Different objectives yield different “best” weights.

  • Short selling

    • Short selling is described as:
      • borrowing an asset and selling it now
      • returning it after the time period ends
    • Consequences in the math/graph:
      • weights can become negative
      • portfolio “lines”/ranges change (dotted lines referenced for short-selling cases).

  • Optimization method (two-asset case)

    • For a two-stock portfolio, optimization is framed by differentiating with respect to a single allocation variable (e.g., (\alpha)).
    • Maximum return
      • Differentiate portfolio return w.r.t. (\alpha)
      • Set derivative to 0
      • Use second derivative sign:
        • “Less than zero” indicates a maximum (as stated).
    • Minimum risk
      • Differentiate portfolio variance w.r.t. (\alpha)
      • Set derivative to 0
      • Second derivative > 0 indicates a minimum (as stated).

  • Diversification insight

    • Uncorrelated assets ((\rho = 0))
      • Correlations are 0 → covariances are 0.
      • Risk reduces as the number of assets increases.
      • With equal weights (w_i = 1/n):
        • expected return stays at the average
        • variance decreases like (\sigma^2/n)
      • In theory, with infinitely many uncorrelated assets, portfolio variance can approach 0.
    • Correlated assets
      • Nonzero covariances mean risk contains a portion that cannot be diversified away.
      • Risk splits conceptually into:
        • diversifiable risk (shrinks with (n))
        • non-diversifiable risk (remains due to correlation)

  • Efficient frontier / feasible set (implied)

    • Varying weights fills a region bounded by extreme cases.
    • Correlation affects the shape and limits of attainable risk/return combinations.
    • Short selling expands the region beyond the “no-short-selling” line segments (dotted lines noted).

  • Risk-free rate link (brief)

    • The y-intercept idea is mentioned:
      • when risk is near zero, it suggests how a risk-free interest rate would be introduced later.

Methodologies / “instructions” presented (bullet format)

A) Build and analyze an n-asset portfolio

Inputs

  • Choose n stocks
  • For each stock (i), compute/estimate:
    • expected return ( \bar{r}_i )
    • variance ( \sigma_i^2 )
  • Determine pairwise:
    • covariances ( \text{Cov}(i,j) ) (or correlations ( \rho_{ij} ))

Define weights

  • Let (w_i) be the fraction invested in stock (i)
  • Impose constraint:
    • normally: (\sum_{i=1}^{n} w_i = 1)
    • with short selling: some (w_i < 0) is allowed, but normalization often still keeps the sum constraint (explained later).

Compute expected portfolio return

  • [ E[R_p] = \sum_{i=1}^{n} w_i \bar{r}_i ]

Compute portfolio variance (risk)

  • Use covariance matrix:

    • [ \sigma_p^2 = \sum_{i=1}^{n}\sum_{j=1}^{n} w_i w_j \text{Cov}(i,j) ]

  • Interpretation:

    • diagonal: (w_i^2\sigma_i^2)
    • off-diagonal: cross-covariance terms (w_i w_j \text{Cov}(i,j))

Optimize for an objective

  • If objective is maximize return:
    • choose weights to maximize (E[R_p])
    • (in the lecture, shown via a derivative method for 2 assets)
  • If objective is minimize risk:
    • choose weights to minimize ( \sigma_p^2 )
    • (again demonstrated with a derivative approach for 2 assets)

B) Two-asset optimization (using calculus with one variable)

Setup

  • Portfolio uses weights (\alpha) and (1-\alpha) across two assets.
  • Portfolio return becomes a function of (\alpha).
  • Portfolio variance becomes a function of (\alpha), including covariance/correlation terms.

To maximize return

  • Differentiate portfolio return w.r.t. (\alpha)
  • Set derivative = 0
  • Use second derivative test to identify a maximum (per the sign rule described).

To minimize risk

  • Differentiate portfolio variance w.r.t. (\alpha)
  • Set derivative = 0
  • Use second derivative test to identify a minimum (described as > 0).

C) Diversification reasoning cases

  • Case 1: uncorrelated assets

    • If correlations are 0, covariances are 0
    • With equal weights (w_i = 1/n):
      • portfolio variance decreases as (1/n)

  • Case 2: correlated assets

    • Correlations imply nonzero covariances
    • A non-diversifiable risk component remains even as (n) grows.

Speakers or sources featured

  • No specific named speaker or external sources are identified in the provided subtitles. (The narration appears to be from a single lecturer, but not named.)

Original video