Applied mathematics / English

Time value of money, Compound interest, Portfolio risk｜Devils-Advocate

Devils-Advocate · Bilibili · 6:38

Bilibili · Devils-AdvocateOriginal video

Selected moment · 0:00 · Chapter markers are time references, not playback progress.

IN THIS VIDEO

01Course topic and prerequisites
02Definition of portfolio return $\mu_{\pi}$
03Expression for Risk Premia

FROM THE LEARNING NOTES

\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)

Definition of portfolio return $\mu_{\pi}$

SELECTED MOMENT

Course introduction and prerequisite requirements

SUGGESTED PREPARATION

calculus
partial differentiation
linear algebra
CAPM model
NumPy matrix operations
Vector and matrix notation
Partial Derivatives and the Hessian
Quadratic Forms
Basic Portfolio Concepts
Optimization with Equality Constraints
Matrix transpose and matrix multiplication
Inverse matrix
Quadratic form
Partial derivative
Method of Lagrange multipliers
Basic notation for mean-variance portfolio
matrix inversion and quadratic forms in linear algebra
partial differentiation of multivariable functions
Lagrange multiplier method
Mean-variance portfolio basic setup
NumPy matrix operation basics

READ & KEEP

The explanation, unpacked.

Saves 0

Reviewed learning material · Video analysis · English

Read the full overview

The first 62 seconds serve as course introduction: the title slide shows 'Mean Variance - Lagrange / Lagrangian Method' and the Chinese text 'Mean-Variance Optimal Portfolio / Lagrange Multiplier Method'; the voiceover mentions Markowitz, Ray Dalio, and Gary Brinson, and lists prerequisites including calculus, partial differentiation, linear algebra, CAPM, and NumPy matrix operations; the background also features Python code and rotating geometric graphics. After 62 seconds, the scene switches to the formula board, first presenting portfolio return $\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)$ ; the voiceover explains the meaning of each term $\mu$ , $w$ , $w^{T}$ , $r_f$ , and $1$ in turn, clarifying that this expression represents the return of a portfolio combining risky and risk-free assets. The visuals then continue to display Risk Premia, the maximization objective function $\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$ , first-order conditions, and second-order conditions—but these later formulas are shown primarily without full derivation or detailed explanation in this segment. This segment first uses a static formula slide to explain mean-variance optimal portfolio selection: write the portfolio return as $μ_p = w^Tμ + r_f(1 - w^T\cdot 1) = r_f + w^T(μ - r_f\cdot 1)$ , pointing out that $μ - r_f\cdot 1$ is the risk premium; then construct the objective max $V(w) = [r_f + w^T(μ - r_f\cdot 1)] - λ/2 w^TΣw$ , explaining that λ represents risk aversion and the variance term acts as a penalty. Next, take the partial derivative with respect to w to obtain the first-order condition ( $μ - r_f\cdot 1$ ) - λΣw^*, and use the Hessian HV(w^*) = -λ $Σ < 0$ to confirm it is a maximum. Approximately 105 seconds later, switch to the Lagrange multiplier slide, presenting the general form $f(x) + Σ_{j=1}^m λ_j(g_j(x)-b_j)$ , and formulate the portfolio problem as minimizing $1/2 w^TΣw$ subject to $w^Tμ = m$ and $w^T1 = 1$ . This 120-second video derives the mean-variance optimal portfolio using the method of Lagrange multipliers. The first half presents the original problem: minimize $1/2 w^TΣw$ subject to $w^Tμ=m$ and $w^T1=1$ ; then constructs the Lagrangian function $L(w,λ,γ)$ , explaining term by term the risk, return constraint, and weight-sum constraint. In the middle part, the first-order condition with respect to w yields $Σw+λμ+γ1=0$ , which is rearranged into w*= $Σ^{-1}(λμ+γ1)$ . Next, substitute w* back into the two constraints, define $A=1^TΣ^{-1}1$ , $B=1^TΣ^{-1}μ$ , $C=μ^TΣ^{-1}μ$ , and solve for λ=(Am-B)/(AC-B^2) and γ=(C-Bm)/(AC-B^2); finally substitute back to obtain the optimal weights. At the end, Python code appears on the right, implementing step by step A, B, C, Lam, Gamma, and weights as in the formulas. This segment explains mean-variance portfolio optimization using the method of Lagrange multipliers, presented in a left-right split layout. On the left: the Lagrangian function, first-order conditions, optimal weight expression, the two-variable system after substituting constraints, and closed-form solutions for the multipliers. On the right: Python code constructing the covariance matrix SRS from standard deviations and correlations, then computing A, B, C, Lam, Gamma, and finally weights. The speaker emphasizes correspondences between code variables and formula symbols—for example, U replaces $\mu$ and SRS corresponds to $\Sigma$ —and summarizes the solution procedure as: first compute A, B, C; then solve for lambda and gamma; finally back-substitute to obtain the optimal portfolio. At the end, the speaker mentions that the derivation and code have been published on the official WeChat account, without showing final numerical outputs.

Use the learning inspector for key ideas and moments, or open the reading tabs for the complete notes.

Chapters

0:00Course introduction and prerequisite requirements 1:02Mean-variance optimization formula board: portfolio return and optimality conditions 2:00Portfolio return and risk premium 2:20Mean-variance objective function 2:55First-order condition 3:26Second-order condition and verification of maximum 3:45Method of Lagrange multipliers and constraint examples 4:00The original optimization problem and its two constraints 4:3.5Construct the Lagrangian function and explain its three terms 4:30Take the first-order condition with respect to w 4:41Solve explicitly for w*4:52Substitute w* back into the constraints 5:16Introduce A, B, and C and solve for λ and γ 5:40Substitute back to obtain the final weights and display the code 6:00SRS and covariance matrix construction 6:08Lagrangian function and first-order conditions 6:15A, B, C and multiplier solving 6:24Back-substitution to obtain optimal weights 6:29Resource Description and Conclusion

Learning script

Generated from the video's visuals and explanation; not verbatim speech.

The opening begins with an investment-related quote to introduce the topic, highlighting the relationship between asset allocation and long-term stable returns, and names Markowitz, Ray Dalio, and Gary Brinson as motivational figures for the course.

Next, the lecture topic is clearly stated as mean-variance optimization and the Lagrange method, with prerequisites listed: mathematically, calculus, partial derivatives, and linear algebra; financially, familiarity with CAPM is recommended; programmatically, proficiency in NumPy matrix operations is recommended.

After moving to the formula board, portfolio return $\mu_{\pi}$ is first defined. The screen splits the return into two parts: $w^{T}\mu$ represents the return from risky assets, and $r_f(1-w^{T}*1)$ represents the return from the risk-free asset; then it is rearranged into $r_f+w^{T}(\mu-r_f1)$ , placing the risk-free return first and writing the remaining part as the product of weights and risk premia.

The voiceover explains each symbol term by term: $\mu$ is the vector of expected returns of assets, $w$ is the weight vector, $w^{T}$ is the transpose of the weight vector, $r_f$ is the risk-free rate, and $1$ is an object composed of 1. You may memorize the formula structure directly, but note that the video’s verbal description of $1$ is relatively brief.

Below on the same page, the video continues to display Risk Premia, the objective function to maximize, the first-order condition, and the second-order condition. The objective function is written as $\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$ , the first-order condition as $\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$ , and the second-order condition as $H V(w^*)=-\lambda\Sigma<0$ , with a note indicating it is used to confirm a maximum. This segment mainly presents these formulas without full derivation or explanation of symbol meanings.

This segment first handles the representation of portfolio return. At the top of the screen appears $\mu_p = w^T\mu + r_f(1 - w^T\mathbf{1}) = r_f + w^T(\mu - r_f\mathbf{1})$ , where $r_f$ is the risk-free return, $w$ is the weight vector of risky assets, and $\mathbf{1}$ is the all-1 vector. After expanding and regrouping the right-hand side, the return is expressed as 'risk-free baseline plus weighted excess return term'.

Next, the video separately names $\mu - r_f\mathbf{1}$ the Risk Premia vector, i.e., the risk premium vector. The key point here is that portfolio return depends not only on absolute expected return but rather on the excess over the risk-free rate.

The next step introduces the objective function. The screen displays $\max V(w) = [r_f + w^T(\mu - r_f\mathbf{1})] - \frac{\lambda}{2}w^T\Sigma w$ . The first term carries forward the return expression from above, while the second term subtracts the portfolio variance as a penalty; the larger $\lambda$ is, the more risk-averse the investor.

Note also a notational difference here: the audio reads the variance term as $w\omega w$ , but the screen actually writes $w^T\Sigma w$ . In this segment, both refer to the same quadratic risk term, though the spoken notation differs from the board notation.

With the objective function established, the video proceeds to find its extremum. Since we maximize $V(w)$ , we take the partial derivative with respect to the weight vector $w$ , yielding the first-order condition $\Delta V(w^*) = \frac{\partial V}{\partial w}(w^*) = (\mu - r_f\mathbf{1}) - \lambda\Sigma w^*$ . This step means that, at the optimal weights $w^*$ , marginal excess return exactly offsets marginal risk penalty.

The video further explains the meaning of $w^*$ : it is not a single number but a vector containing the weights of all assets; if there are four assets, $w^*$ contains four components.

The first-order condition yields only a stationary point; we must still determine whether it is a maximum or a minimum. Thus, the screen next presents the second-order condition $HV(w^*) = -\lambda\Sigma < 0$ , labeled 'confirms we found Maximum'. That is, when the Hessian is negative definite, the stationary point corresponds to a maximum.

Here the audio uses the phrases 'concave down' and 'concave up', which are intuitive analogies for univariate functions; strictly speaking, for the multivariate optimization in this segment, this corresponds to the negative definiteness test of the Hessian.

The slide then switches to the Lagrange Method. The new formula page first presents the general Lagrangian function $f(x) + \sum_{j=1}^{m}\lambda_j(g_j(x) - b_j)$ , where the summation term is labeled 'Penalty', indicating that the multiplier term enforces constraint penalties.

Next, the video applies this general form to portfolio optimization: the objective is written as $\min(w): \frac{1}{2}w^T\Sigma w$ , annotated beside it as 'risk'; the constraints are written as $w^T\mu = m$ and $w^T\mathbf{1} = 1$ , representing respectively that the portfolio return equals target $m$ and that the sum of all weights equals 1. The segment ends here, without proceeding to solve for the multipliers or optimal weights.

At the beginning, the video clearly states the problem: this is a mean-variance optimization problem, aiming to minimize portfolio risk $\frac{1}{2}w^{T}\Sigma w$ while satisfying two constraints—one requiring the portfolio’s expected return to equal $w^{T}\mu=m$ , and the other requiring the sum of all asset weights to equal $w^{T}\mathbf{1}=1$ . Here, $w$ is the unknown asset weight vector, $\Sigma$ is the covariance matrix, $\mu$ is the expected return vector, and $m$ is the target return.

Next, the screen expands the original problem into the Lagrangian function $L(w,\lambda,\gamma)=\frac{1}{2}w^{T}\Sigma w+\lambda(m-w^{T}\mu)+\gamma(1-w^{T}\mathbf{1})$ . The presenter explains in order that these three terms correspond respectively to the risk term, the return constraint term, and the weight-sum constraint term; thus $\lambda$ and $\gamma$ are the multipliers associated with the two constraints.

Then proceed to the first solution step: take the partial derivative of $w$ with respect to its variables. The first-order condition $\frac{\partial L(w,\lambda,\gamma)}{\partial w}=\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$ appears on screen. This step converts the question "what condition should the optimal weights satisfy?" from a functional form into an equation form.

Within this equation, the video continues with algebraic manipulation to isolate $w$ , yielding $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ . This shows that if the multipliers $\lambda$ and $\gamma$ are known, the optimal weights can be written down directly; the problem thus reduces to determining these two multipliers.

Next, substitute $w^*$ back into the original two constraints. The return constraint becomes $\mu^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=m$ , and the weight-sum constraint becomes $\mathbf{1}^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=1$ . After expansion, this yields a linear system of two equations in the two unknowns $\lambda$ and $\gamma$ .

To make the system more visually appealing, the video introduces three scalar quantities: $A=\mathbf{1}^{T}\Sigma^{-1}\mathbf{1}$ , $B=\mathbf{1}^{T}\Sigma^{-1}\mu$ , and $C=\mu^{T}\Sigma^{-1}\mu$ . The two preceding equations then compress into $\lambda C+\gamma B=m$ and $\lambda B+\gamma A=1$ . This step introduces no new conclusion—it merely renames complex expressions to simplify solving.

The solution is then given directly: $\lambda=\frac{Am-B}{AC-B^{2}}$ and $\gamma=\frac{C-Bm}{AC-B^{2}}$ . The video does not demonstrate the elimination steps explicitly, but the formula structure indicates these are closed-form solutions derived from the two-variable linear system, requiring the denominator $AC-B^{2}$ to be nonzero.

Finally, the instructor emphasizes that the true objective is not the multipliers themselves, but the weight vector $w$ . Substituting the computed $\lambda$ and $\gamma$ back into $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ yields the weight assigned to each asset in the optimal portfolio. Simultaneously, Python code appears on the right, sequentially constructing `SRS`, `ones`, then computing `A`, `B`, `C`, `Lam`, `Gamma`, and `weights`, translating each left-hand-side formula into executable code.

The video opens with a side-by-side layout: on the left is 'Lagrange Method', and on the right is a segment of Python code. The presenter first points to 'SRS' in the code and says it equals sigma times correlation times the transpose of sigma. The mathematical meaning is clear: the covariance matrix $\Sigma$ in the formula is named 'SRS' in the code, and it is constructed from the vector of standard deviations and the correlation matrix.

The camera remains on the same slide; on the left, the Lagrangian function $L(w,\lambda,\gamma)=\frac12 w^T\Sigma w+\lambda(m-w^T\mu)+\gamma(1-w^T\mathbf{1})$ is visible, followed below by the first-order conditions $\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$ , and then $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ . The presenter adds that variables B and C in the code are computed analogously to A, and that $\mu$ in the formula is replaced by U in the code. This segment formally aligns theoretical symbols with programming variables.

Next, the presenter outlines the solution order: compute A, B, and C first, then substitute them into lambda and gamma. On screen, the left side has already substituted the weight expression back into the two constraints and simplified them into $C\lambda+B\gamma=m$ and $B\lambda+A\gamma=1$ ; the right side computes A, B, and C separately using matrix multiplication. The reasoning is that substituting $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ into the return constraint and the budget constraint yields a linear system of equations in only two unknown multipliers.

Then the screen displays closed-form solutions for the multipliers $\lambda=\frac{Am-B}{AC-B^2}$ and $\gamma=\frac{C-Bm}{AC-B^2}$ ; the corresponding code on the right defines them as 'Lam' and 'Gamma'. The presenter states that finally plugging lambda and gamma into 'weights' yields the optimal portfolio. This step connects to the bottommost formula $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ : once the two multipliers are determined, the optimal weights can be computed directly.

In the final ten seconds, no new derivations appear; the screen remains the same split-layout image, with the watermark 'WeChat Official Account: Devil's Spokesperson' appearing at the bottom. The presenter explains that both the derivation and the code have been posted to the WeChat official account, and thanks viewers. Mathematically, this segment concludes the full solution chain: starting from the Lagrangian function, proceeding through first-order conditions, substitution into constraints, solving for multipliers, and ending with code implementation of the optimal weights.

Knowledge cards

Course topic and prerequisites

At the beginning of this segment, the topic is set as 'mean-variance optimal portfolio' and 'Lagrange multiplier method'. The voiceover introduces the importance of asset allocation first, then lists learning prerequisites: mathematically, calculus, partial derivatives, and linear algebra; financially, understanding CAPM is recommended; programmatically, familiarity with NumPy’s matrix operations is recommended. This card corresponds to the introduction and does not involve specific derivations.

Definition of portfolio return $\mu_{\pi}$

The screen displays $\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)$ . The voiceover explains that $\mu$ is the vector of expected returns of risky assets, $w$ is the weight vector, $w^{T}$ is the transpose of the weight vector, $r_f$ is the risk-free rate, and $1$ is an object composed of 1. The first half of the formula splits the return into the risky-asset part and the risk-free-asset part; the second half rearranges it into the form of risk-free return plus the weighted risk premia.

\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)

Expression for Risk Premia

This segment isolates $(\mu-r_f*1)$ as Risk Premia. It represents the vector formed by subtracting the risk-free rate from the expected returns of each risky asset, and appears directly in the rearranged portfolio return expression and subsequent objective function. The video does not provide further verbal explanation of its economic meaning within this segment.

\mu-r_f*1

Display of the objective function to maximize

The screen displays $\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$ . It can be directly confirmed that the objective function consists of the portfolio return term minus a quadratic form term involving $\lambda$ and $\Sigma$ . This segment does not explain the meanings of $V(w)$ , $\lambda$ , or $\Sigma$ , nor does it specify the full constraints of the maximization problem.

\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w

First-order and second-order conditions

On the same formula board, the video continues to display $\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$ and $H V(w^*)=-\lambda\Sigma<0$ , with the latter annotated "confirms we found Maximum". This indicates that the visual layout presents optimality criteria in the order "objective function—first-order conditions—second-order conditions", but this segment does not show the derivation from the objective function to these two conditions, nor does it explain the precise meaning of the matrix inequality.

\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*,\quad H V(w^*)=-\lambda\Sigma<0

Portfolio return with a risk-free asset

The video expresses portfolio return as a weighted sum of risky-asset returns and the risk-free return, then rewrites it in a form benchmarked against the risk-free rate.

\mu_p = w^T\mu + r_f(1 - w^T\mathbf{1}) = r_f + w^T(\mu - r_f\mathbf{1})

Risk premium vector

In the rewritten portfolio return expression, $\mu - r_f\mathbf{1}$ is defined as the risk premium, representing the excess expected return of each asset over the risk-free rate.

\mu - r_f\mathbf{1}

Mean-variance optimization objective function

The objective function combines return maximization with a variance penalty: the return term comes from the earlier portfolio return expression, the risk term is $\frac{\lambda}{2}w^T\Sigma w$ , and $\lambda$ measures the degree of risk aversion.

\max V(w) = \left[r_f + w^T(\mu - r_f\mathbf{1})\right] - \frac{\lambda}{2}w^T\Sigma w

First-order condition

Taking the partial derivative of $V(w)$ with respect to the weights $w$ and setting it to zero yields the condition satisfied by the optimal weights $w^*$ .

\Delta V(w^*) = \frac{\partial V}{\partial w}(w^*) = (\mu - r_f\mathbf{1}) - \lambda\Sigma w^*

Second-order condition and maximum verification

The video uses the Hessian to determine the nature of the stationary point, gives $HV(w^*) = -\lambda\Sigma < 0$ , and thereby confirms that the solution found is a maximum.

HV(w^*) = -\lambda\Sigma < 0

General form of the Lagrangian function

An optimization problem with $m$ equality constraints can be written as the sum of the objective function and multiplier penalty terms.

f(x) + \sum_{j=1}^{m}\lambda_j(g_j(x) - b_j)

Risk minimization and portfolio constraints

In the Lagrangian example, the goal is to minimize semivariance risk while requiring the portfolio return to equal $m$ and the sum of weights to equal 1.

\min(w): \frac{1}{2}w^T\Sigma w,\quad \text{s.t. } w^T\mu = m,\; w^T\mathbf{1} = 1

The original mean-variance optimization problem

The video first presents the optimization problem to be solved: minimize portfolio risk subject to target-return and budget constraints. The objective function is $\frac{1}{2}w^{T}\Sigma w$ ; the two constraints are $w^{T}\mu=m$ and $w^{T}\mathbf{1}=1$ . Here, $w$ is the asset weight vector, $\Sigma$ is the covariance matrix, $\mu$ is the expected return vector, $m$ is the target portfolio return, $\mathbf{1}$ is the all-1 vector.

\min_{w}\;\frac{1}{2}w^{T}\Sigma w\quad\text{s.t.}\quad w^{T}\mu=m,\;w^{T}\mathbf{1}=1

The three terms of the Lagrangian function

To convert the constrained optimization into a differentiable problem, the video constructs the Lagrangian function $L(w,\lambda,\gamma)=\frac{1}{2}w^{T}\Sigma w+\lambda(m-w^{T}\mu)+\gamma(1-w^{T}\mathbf{1})$ . The first term represents risk; the second term corresponds to the return constraint; the third term corresponds to the weight-sum constraint; $\lambda$ and $\gamma$ are the multipliers associated with those constraints.

L(w,\lambda,\gamma)=\frac{1}{2}w^{T}\Sigma w+\lambda(m-w^{T}\mu)+\gamma(1-w^{T}\mathbf{1})

First-order condition with respect to weights

The optimal weights must set the partial derivative of the Lagrangian function with respect to $w$ equal to zero. The video writes down $\frac{\partial L(w,\lambda,\gamma)}{\partial w}=\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$ . This step transforms the original problem into a linear equation.

\frac{\partial L(w,\lambda,\gamma)}{\partial w}=\Sigma w+\lambda\mu+\gamma\mathbf{1}=0

Explicit expression for the optimal weights

Rearranging the first-order condition and left-multiplying by $\Sigma^{-1}$ yields $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ . This means that once the two multipliers are known, the optimal weights can be computed directly.

w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})

Substituting w* back into the constraints

To determine $\lambda$ and $\gamma$ , the video substitutes $w^*$ back into the original two constraints, obtaining $\mu^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=m$ and $\mathbf{1}^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=1$ . After expansion, this becomes a system of two linear equations in $\lambda$ and $\gamma$ .

\begin{cases}\mu^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=m\\\mathbf{1}^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=1\end{cases}

Three intermediate quantities: A, B, and C

The video denotes scalars that repeatedly appear after substituting into the constraints as $A=\mathbf{1}^{T}\Sigma^{-1}\mathbf{1}$ , $B=\mathbf{1}^{T}\Sigma^{-1}\mu$ , and $C=\mu^{T}\Sigma^{-1}\mu$ . Thus, the system of equations can be written concisely as $\lambda C+\gamma B=m$ and $\lambda B+\gamma A=1$ .

A=\mathbf{1}^{T}\Sigma^{-1}\mathbf{1},\quad B=\mathbf{1}^{T}\Sigma^{-1}\mu,\quad C=\mu^{T}\Sigma^{-1}\mu

Closed-form solution for the Lagrange multipliers

From the simplified two-variable system, the video directly gives $\lambda=\frac{Am-B}{AC-B^{2}}$ and $\gamma=\frac{C-Bm}{AC-B^{2}}$ . This step requires $AC-B^{2}\neq 0$ .

\lambda=\frac{Am-B}{AC-B^{2}},\qquad \gamma=\frac{C-Bm}{AC-B^{2}}

Final formula for the optimal weights

Finally, the computed $\lambda$ and $\gamma$ are substituted back into $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ to obtain the weight of each asset in the optimal portfolio. The video emphasizes that the final goal is $w$ , not the intermediate multipliers.

w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})

Correspondence between the formula and Python code

The code on the right implements the analytical derivation term by term: first constructing `SRS` using `sigma` and `correlation`, then using `ones` to represent the all-1 vector, followed by computing `A`, `B`, `C`, `Lam`, and `Gamma`, and finally obtaining the optimal weights via `weights = np.matrix(SRS).I * (Lam*u+Gamma*ones)`.

weights = np.matrix(SRS).I * (Lam*u+Gamma*ones)

Setting Up the Lagrangian Function

The video formulates mean-variance optimization as a Lagrangian problem with two equality constraints: minimize $\frac12 w^T\Sigma w$ subject to the expected return constraint $m-w^T\mu=0$ and the budget constraint $1-w^T\mathbf{1}=0$ . The corresponding multipliers are $\lambda$ and $\gamma$ .

L(w,\lambda,\gamma)=\frac12 w^T\Sigma w+\lambda(m-w^T\mu)+\gamma(1-w^T\mathbf{1})

First-Order Conditions and the Weight Expression

Taking the derivative of $w$ and setting the gradient to zero yields $\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$ . If $\Sigma$ is invertible, left-multiplying by $\Sigma^{-1}$ gives the explicit form of the optimal weights $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ .

\Sigma w+\lambda\mu+\gamma\mathbf{1}=0,\quad w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})

Substituting into Constraints to Obtain a Two-Variable System

Substituting $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ back into the two original constraints, the video derives a linear system in terms of $\lambda,\gamma$ . For brevity, scalar coefficients $A=\mathbf{1}^T\Sigma^{-1}\mathbf{1}$ , $B=\mu^T\Sigma^{-1}\mathbf{1}$ , and $C=\mu^T\Sigma^{-1}\mu$ are introduced, so the system becomes $C\lambda+B\gamma=m$ and $B\lambda+A\gamma=1$ .

\begin{cases} C\lambda+B\gamma=m \\ B\lambda+A\gamma=1 \end{cases}

Detailed learning notes

Explore conditions, steps and evidence. Supplementary explanations are labeled separately from content shown in the video.

Symbols · 62

$\mu_{\pi}$

Clear evidence

Shown in the video

Evidence

Formula
Observation
Below the label 'Portfolio Return' appears $\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)$
Audio
Observation
The narrator says, "the portfolio's return, that is, mu pi here"

Symbol

$\mu_{\pi}$

Meaning

portfolio return

Domain

scalar return quantity; the video does not specify the unit

$w$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The formula $\mu_{\pi}$ contains $w^{T}\mu$ and $(1-w^{T}*1)$
Audio
Observation
The narrator says, "and w is our weight vector"

Symbol

$w$

Meaning

weights vector for risky assets

Domain

The video does not explicitly state the dimension; from matrix operations, it appears as a column vector participating in a transpose operation

$w^{T}$

Clear evidence

Shown in the video

Evidence

Formula
Observation
$w^{T}$ appears repeatedly in the formula
Audio
Observation
The narrator says, "W to the power of T, that is, W superscript T, denotes the transpose of this matrix, namely, the transpose of its weights"

Symbol

$w^{T}$

Meaning

transpose of the weights vector

Domain

row vector; used for matrix multiplication with a column vector

$\mu$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The formula $\mu_{\pi}$ contains $w^{T}\mu$ and $(\mu-r_f1)$
Audio
Observation
The narrator says, "The mu after the equals sign represents the expected return of each asset in this portfolio."

Symbol

$\mu$

Meaning

Vector of expected returns of the risky assets

Domain

Vector; the video does not specify its dimension

$r_f$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The formula $\mu_{\pi}$ contains $r_f(1-w^{T}*1)$ and $r_f+w^{T}(\mu-r_f1)$
Audio
Observation
The narrator says, "Rf is the risk-free rate, i.e., the risk-free return rate."

Symbol

$r_f$

Meaning

Risk-free return rate

Domain

Scalar; the video does not specify its unit

$1$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The formula contains $1$ , such as $(1-w^{T}*1)$ and $(\mu-r_f1)$
Audio
Observation
The narrator says, "One of them is a matrix consisting entirely of ones."

Uncertainties

The narrator calls it a "matrix", but in the on-screen formula this symbol serves as the all-1 vector; the video does not further distinguish between vectors and matrices.

Symbol

$1$

Meaning

Notation for the vector/matrix consisting entirely of 1

Domain

The video does not specify the dimension; used in formulas together with $w^{T}$ , $\mu$ , and $r_f$

$(\mu-r_f*1)$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays Risk Premia as $(\mu-r_f*1)$

Symbol

$(\mu-r_f*1)$

Meaning

Risk premium vector

Domain

Vector; the video does not specify its dimension

$V(w)$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays $\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$

Uncertainties

The name of $V(w)$ is not explained in the audio; the video does not state whether it is a utility function.

Symbol

$V(w)$

Meaning

Objective function

Domain

Scalar function of $w$ ; the video does not specify its domain or range

$\lambda$

Clear evidence

Shown in the video

Evidence

Formula
Observation
$\frac{\lambda}{2}w^{T}\Sigma w$ appears in the objective function; $-\lambda\Sigma$ appears in the second-order condition

Uncertainties

The video does not explain the meaning, range of values, or economic interpretation of $\lambda$ .

Symbol

$\lambda$

Meaning

Parameter in the objective function

Domain

Scalar; the video does not specify its domain

$\Sigma$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The objective function contains $w^{T}\Sigma w$ ; the second-order condition contains $-\lambda\Sigma$

Uncertainties

The video does not explain the meaning of $\Sigma$ ; from the notation, it is typically the covariance matrix, but this is not directly stated in this segment.

Symbol

$\Sigma$

Meaning

The matrix symbol appearing in the objective function

Domain

A matrix; the video does not specify its dimensions or properties

$w^*$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The first-order condition is written as $\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$

Uncertainties

The video does not explain how $w^*$ is obtained, nor does it state whether it is unique.

Symbol

$w^*$

Meaning

The point discussed in the first-order and second-order conditions

Domain

A vector; the video does not specify its dimension

$H V(w^*)$

Clear evidence

Shown in the video

Evidence

Formula
Observation
The on-screen title reads Second order condition (Hessian), with $H V(w^*)=-\lambda\Sigma<0$ given below

Uncertainties

The video does not explain the definition of the Hessian; it appears only as a title and formula.

Symbol

$H V(w^*)$

Meaning

The Hessian notation of the objective function at $w^*$

Domain

A matrix; the video does not specify its dimensions

Knowledge points · 30

Mean-Variance Portfolio Optimization

Clear evidence

Shown in the video

Evidence

Formula
Observation
Title at the top of the screen: Mean Variance Optimization / Mean-Variance Portfolio Optimization

Definition

Explanation

The second half of this segment is titled 'Mean Variance Optimization / Mean-Variance Portfolio Optimization' and focuses on portfolio return, risk premium, and an objective function containing term $w^{T}\Sigma w$ . The video directly presents the title and the formula framework but does not fully explain the meaning of 'mean-variance' within this segment.

Conditions

Only the title and formula are shown in the segment; no complete definition is expanded.

Portfolio return formula

Clear evidence

Shown in the video

Evidence

Formula
Observation
Under Portfolio Return: $\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)$
Audio
Observation
The voiceover explains the meanings of $\mu_{\pi}$ , $\mu$ , $w$ , $w^{T}$ , $r_f$ , and $1$ one by one, and states that this portfolio consists of risky assets and a risk-free asset.

Formula

Explanation

The screen shows two equivalent expressions for portfolio return $\mu_{\pi}$ : first written as the sum of risky-asset returns and risk-free-asset return, then rearranged as the risk-free return plus the inner product of weights and risk premia. The voiceover explicitly states that $\mu$ denotes the expected values of the assets, $w$ denotes the weights, $w^{T}$ denotes the transpose of the weights, $r_f$ denotes the risk-free rate of return, and $1$ denotes the matrix/vector notation composed of 1.

Formula

\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)

Conditions

The video does not specify the dimensions of $w$ , $\mu$ , or $1$ .
The video does not state whether the weights satisfy any constraints other than summing to 1.

Prerequisites

$\mu_{\pi}$
$w$
$w^{T}$
$\mu$
$r_f$
$1$

Risk premium expression

Clear evidence

Shown in the video

Evidence

Formula
Observation
Under Portfolio Return, Risk Premia is listed separately as $(\mu-r_f*1)$ .

Uncertainties

The video does not verbally explain the economic definition of 'Risk Premia'; it is given only in formula form.

Definition

Explanation

The segment labels $(\mu-r_f*1)$ as Risk Premia, i.e., the vector form of each risky asset's expected return minus the risk-free rate. This expression then enters the rearranged portfolio return formula and the objective function.

Formula

\mu-r_f*1

Conditions

The video does not explain the dimension-matching details between $\mu$ and $1$ .

Prerequisites

$\mu$
$r_f$
$1$
$(\mu-r_f*1)$

Optimization objective function

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays: In a market with N risky assets and a risk-free asset: $\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$
Audio
Observation
The audio in this segment does not explain this objective function; it only explains the portfolio return formula shown above.

Uncertainties

The meanings of $V(w)$ , $\lambda$ , and $\Sigma$ are not verbally explained within this segment.
The English sentence on screen ends with "risk-free as:", suggesting truncation or incomplete typesetting.

Formula

Explanation

The screen presents a maximization problem: the objective function $V(w)$ equals the portfolio return term $r_f+w^{T}(\mu-r_f1)$ minus $\frac{\lambda}{2}w^{T}\Sigma w$ . The video directly displays the formula itself but does not explain, within this segment, why it is constructed this way, the role of $\lambda$ , or the specific meaning of $\Sigma$ .

Formula

\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w

Conditions

The video does not state any assumptions beyond $N$ risky assets and one risk-free asset.
The video does not specify the constraints of the maximization problem.

Prerequisites

Portfolio return formula
Risk premium expression
$V(w)$
$\lambda$
$\Sigma$

First-order condition

Clear evidence

Shown in the video

Evidence

Formula
Observation
Under "First order condition", the screen displays $\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$ .

Uncertainties

The video does not show the derivation of this first-order condition.
It does not explain how $w^*$ is obtained.

Formula

Explanation

The segment directly gives the first-order condition of the objective function at $w^*$ , in the form that the gradient equals $(\mu-r_f1)-\lambda\Sigma w^*$ . The screen displays $\Delta V(w^*)$ and $\frac{\partial V}{\partial w}(w^*)$ side by side, indicating that the discussion here concerns the first derivative or gradient condition with respect to $w$ .

Formula

\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*

Conditions

The video does not specify whether this condition corresponds to a maximum, a minimum, or merely a stationary point.
The video does not provide the differentiation steps.

Prerequisites

Optimization objective function
$w^*$
$(\mu-r_f*1)$
$\lambda$
$\Sigma$

Second-order condition and maximum determination

Clear evidence

Shown in the video

Evidence

Formula
Observation
Under the heading 'Second order condition (Hessian)', the screen displays $H V(w^*)=-\lambda\Sigma<0$ , accompanied by the note '(confirms we found Maximum)'.
Audio
Observation
The audio in this segment does not explain the second-order condition.

Uncertainties

The video does not explain the definition of the Hessian.
It does not explain why $-\lambda\Sigma<0$ confirms a maximum, nor does it state the assumptions about the properties of $\lambda$ and $\Sigma$ .

Formula

Explanation

The screen presents the second-order condition $H V(w^*)=-\lambda\Sigma<0$ and notes beside it that this is used to confirm that a maximum has been found. The segment directly shows the conclusion without providing the full premises of the relevant theorem.

Formula

H V(w^*)=-\lambda\Sigma<0

Conditions

The video does not specify the precise meaning of the matrix inequality.
The video does not specify what conditions $\lambda$ and $\Sigma$ must satisfy.

Prerequisites

First-order condition
$H V(w^*)$
$\lambda$
$\Sigma$
$w^*$

portfolio return with a risk-free asset

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays Portfolio Return: $μ_p = w^T μ + r_f(1 - w^T \cdot 1) = r_f + w^T(μ - r_f 1)$ .
Audio
Observation
The audio explains that $r_f$ is the risk-free asset return and rewrites the expression as $r_f$ plus w times μ minus the risk-free rate times 1.

Formula

Explanation

The video first writes the portfolio return as the sum of contributions from risky asset weights and the risk-free asset weight, then isolates the risk-free part to obtain a form centered at $r_f$ and weighted by the excess return vector.

Formula

\mu_p = w^T\mu + r_f(1 - w^T\mathbf{1}) = r_f + w^T(\mu - r_f\mathbf{1})

Conditions

There are N risky assets and one risk-free asset
w denotes the vector of weights on risky assets
$\mathbf{1}$ is the all-1 vector

risk premium vector

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen labels Risk Premia ( $μ - r_f \cdot 1$ ) below the portfolio return expression.
Audio
Observation
The audio states that μ minus the risk-free rate times 1 is the risk premium and notes that those familiar with CAPM will recognize this as the risk premium.

Definition

Explanation

The video defines $μ - r_f\cdot 1$ as the risk premium, i.e., the vector of excess expected returns of the risky assets over the risk-free rate.

Formula

\mu - r_f\mathbf{1}

Conditions

Used to express the portfolio return including the risk-free asset.

Prerequisites

portfolio return with a risk-free asset

Mean-variance optimization objective function.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays max $V(w) = [r_f + w^T(μ - r_f 1)] - λ/2 w^TΣw$ .
Audio
Observation
The audio states that, later, we subtract lambda divided by 2 multiplied by w omega w; lambda measures the degree of risk aversion, and w omega w represents the variance, serving as a penalty term for risk.

Uncertainties

The audio pronounces the covariance matrix as 'omega', while the screen writes it as Σ.

Formula

Explanation

The video constructs the maximization objective $V(w)$ : the first term is the portfolio return, and the second term is a penalty on portfolio variance weighted by $λ/2$ ; the larger λ is, the stronger the reduction in the objective due to risk.

Formula

\max V(w) = \left[r_f + w^T(\mu - r_f\mathbf{1})\right] - \frac{\lambda}{2}w^T\Sigma w

Conditions

N risky assets and one risk-free asset.
λ is the coefficient of risk aversion.
Σ is the covariance matrix.

Prerequisites

portfolio return with a risk-free asset
risk premium vector

First-order condition and meaning of optimal weights

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays the first-order condition and ΔV(w^*) = ∂V/∂w(w^*) = ( $μ - r_f 1$ ) - λΣw^*.
Audio
Observation
The audio states that to find the maximum, we differentiate it using the partial derivative of V with respect to w, and notes that w represents our weights.

Method

Explanation

The video takes the partial derivative of $V(w)$ with respect to the weight vector w and sets it to zero, yielding the first-order condition satisfied by the optimal weight w^*; w^* represents the portfolio weight allocation that maximizes return.

Formula

\Delta V(w^*) = \frac{\partial V}{\partial w}(w^*) = (\mu - r_f\mathbf{1}) - \lambda\Sigma w^*

Conditions

Differentiate with respect to w
w^* is a candidate for the optimal weight

Prerequisites

Mean-variance optimization objective function.

Second-order condition determines a maximum

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays Second order condition (Hessian) and HV(w^*) = -λ $Σ < 0$ , with a note confirming we found a maximum.
Audio
Observation
The audio states that the second derivative test is required to verify whether a maximum or minimum has been found: a downward-opening shape yields a maximum, an upward-opening shape yields a minimum; finally, since the result is less than zero, finding a maximum is confirmed.

Method

Explanation

The video uses the Hessian to determine the nature of the critical point, gives HV(w^*) = -λ $Σ < 0$ , and thereby confirms that a maximum has been found.

Formula

HV(w^*) = -\lambda\Sigma < 0

Conditions

λ is greater than 0
Σ is positive definite or at least makes -λΣ negative definite

Prerequisites

First-order condition and meaning of optimal weights

Lagrangian function form

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays Lagrangian Function: $f(x) + Σ_{j=1}^m λ_j(g_j(x) - b_j)$ , with the annotation (Penalty) beneath the summation term.
Audio
Observation
The audio states that next we examine the method of Lagrange multipliers, where the previous formula is denoted by $f(x)$ , and the subsequent terms represent its constraints.

Formula

Explanation

The video introduces the Lagrangian function for constrained optimization, combining the original objective $f(x)$ with a weighted sum of constraint deviations $g_j(x) - b_j$ ; the weights $λ_j$ serve as penalties.

Formula

f(x) + \sum_{j=1}^{m}\lambda_j(g_j(x) - b_j)

Conditions

There are m equality constraints.

Claims and conditions · 9

first-order stationary point condition

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays First order condition: $\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$ .

Uncertainties

The video does not specify the full name of the theorem corresponding to this condition or its assumptions.
It is not stated whether conditions beyond differentiability are required.

Proposition

Statement

Under the notation given in the clip, the first-order condition for the objective function $V(w)$ at $w^*$ is written as $\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$ .

Hypotheses

There exists an objective function $V(w)$ .
can be at $w^*$ to take partial derivatives or gradient of $w$ take the partial derivative/gradient
The video does not specify other assumptions.

Quantifiers

Regarding $w^*$ appearing in the clip and the given symbol $\mu,r_f,\lambda,\Sigma$ .

The second-order condition confirms the maximum.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays Second order condition (Hessian): $H V(w^*)=-\lambda\Sigma<0$ .
Formula
Observation
A note beside confirms we found the maximum.

Uncertainties

The video does not provide all prerequisites needed to determine the maximum.
The precise definition of the matrix inequality $<0$ is not explained.

Proposition

Statement

The segment claims that if the second-order condition $H V(w^*)=-\lambda\Sigma<0$ holds, then we can confirm having found the maximum.

Hypotheses

We have discussed the second-order condition for the objective function $V(w)$ at $w^*$ .
The video does not specify the assumptions about the properties of $\lambda$ and $\Sigma$ .
The video does not clarify whether the first-order condition must be used together with this one.

Quantifiers

Regarding $w^*$ appearing in the segment,

Negative definite Hessian confirms a maximum.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays HV(w^*) = -λ $Σ < 0$ , with the note 'confirms we found Maximum'.
Audio
Observation
The audio states that if the parabola opens downward, the point we find is a maximum; the final result being less than zero confirms that finding the maximum is valid.

Uncertainties

The video does not explicitly state the positive definiteness assumptions on λ and Σ required for this conclusion.

Proposition

Statement

If the second-order condition yields HV(w^*) = -λ $Σ < 0$ , then the critical point w^* corresponds to a maximum of the objective function.

Hypotheses

We have obtained the first-order condition critical point w^*.
λ is greater than 0.
Σ makes -λΣ negative definite.

Quantifiers

Holds for the critical point w^* of the given objective function $V(w)$ .

Oral connection between risk premium and CAPM

Approximate timing

Shown in the video

Evidence

Audio
Observation
The audio states that, if familiar with the CAPM model, this is its risk premium.
Formula
Observation
The screen labels Risk Premia as ( $μ - r_f \cdot 1$ ).

Uncertainties

The video only draws an oral analogy and does not elaborate on the formal definition or derivation of CAPM.

Proposition

Statement

The video calls $μ - r_f\cdot 1$ the risk premium and orally notes that, for those familiar with CAPM, this is precisely its risk premium.

Hypotheses

Expression of portfolio return involving the risk-free rate

Quantifiers

Serves only as an analogy in the video.

Meaning of the two constraints

Clear evidence

Shown in the video

Evidence

Formula
Observation
On screen: Subject to: $w^Tμ=m$ , $w^T1=1$ .
Caption evidence
Observation
The green annotation explains that m is the portfolio return and the sum of all weights equals 1.

Proposition

Statement

In mean-variance optimization, $w^Tμ=m$ means the portfolio’s expected return equals the target value m, and $w^T1=1$ means the sum of all asset weights equals 1.

Hypotheses

w is the weight vector
μ is the expected return vector
1 is the all-1 vector

Quantifiers

holds for all w satisfying the problem assumptions.

What do the three terms of the Lagrangian function represent?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The instructor explicitly states that this formula has three parts: the first part represents risk, the second part represents portfolio return, and the third part represents the constraint that weights sum to 1.
Formula
Observation
The screen displays the three terms correspondingly.

Proposition

Statement

The three terms of $L(w,λ,γ)$ correspond respectively to the portfolio risk term, the return constraint penalty term, and the weight-sum constraint penalty term.

Hypotheses

Use the form of $L(w,λ,γ)$ given in the video.

Quantifiers

The Lagrangian function for this optimization problem.

The purpose of introducing A, B, and C

Clear evidence

Shown in the video

Evidence

Audio
Observation
The instructor says that, to simplify solving this formula, the more complex parts are denoted separately as A, B, and C.
Formula
Observation
The screen gives the definitions of A, B, and C.

Proposition

Statement

Introducing A, B, and C compresses the more complex expression obtained after substituting the constraint into a more easily solvable two-variable linear system.

Hypotheses

Having obtained 〔w^*〕, substitute it back into the constrained expression.

Quantifiers

During this derivation.

The ultimate quantity to solve for.

Clear evidence

Shown in the video

Evidence

Audio
Observation
The presenter states that 〔w〕 is the weight of each asset in our optimal portfolio, and 〔w〕 is the final result we seek.

Proposition

Statement

The ultimate goal of this derivation is to find the weight vector 〔w〕 for each asset in the optimal portfolio.

Hypotheses

The problem is mean-variance optimization.

Quantifiers

Regarding this optimization problem.

Once the multipliers are found, the optimal portfolio weights can be recovered by substitution.

Clear evidence

Shown in the video

Evidence

Audio
Observation
The speaker says: After computing A, B, and C, substitute them into lambda and gamma… Finally, we substitute lambda and gamma into weights to obtain our optimal portfolio.
Formula
Observation
The left side simultaneously displays the formulas for $\lambda$ and $\gamma$ , and w^*= $\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ .

Proposition

Statement

If $\lambda$ and $\gamma$ have been obtained, the optimal weights can be directly computed using $w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$ .

Hypotheses

The first-order conditions hold.
Sigma is invertible
lambda and gamma are determined by the constraint system

Quantifiers

holds for the optimal solution given m, mu, and Sigma

Derivations and proofs · 12

Rewriting the portfolio return formula

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays $\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)=r_f+w^{T}(\mu-r_f1)$ on the same line.
Audio
Observation
The voice-over explains that $w^{T}\mu$ represents the return of the risky asset, $(1-w^{T}*1)$ corresponds to the weight portion of the risk-free asset, and then multiplies by $r_f$ .

Uncertainties

The video does not write out the algebraic expansion step by step; it only shows the initial and equivalent final forms.

Intuitive argument

Steps

Expression
$\mu_{\pi}=w^{T}\mu+r_f(1-w^{T}*1)$
Explanation
The screen first writes portfolio return as the sum of the risky asset portion and the risk-free asset portion.
Justification
This comes directly from the video's formula and the narrator's explanation of the meaning of each term.
Shown in the video
Expression
$=r_f+w^{T}(\mu-r_f1)$
Explanation
On the same line, it continues with the rearranged equivalent form: the constant term $r_f$ is moved to the front, and the remaining part is written as the product of weights and risk premium.
Justification
The video directly shows two expressions connected by an equals sign; no intermediate algebraic steps are shown in this segment.
Shown in the video

Conclusion

This segment shows the equivalent rewriting of portfolio return—from 'risky asset return plus risk-free asset return' to 'risk-free return plus weighted risk premium'.

Displaying the transition from the objective function to the first- and second-order conditions.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen lists the objective function, first-order condition, and second-order condition in sequence.
Audio
Observation
The audio in this segment does not explain these subsequent formulas.

Uncertainties

The video does not show the derivation from the objective function to the first-order condition, nor from there to the second-order condition.
Matrix differentiation rules are not explained.

Visual argument

Steps

Expression
$\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$
Explanation
The screen first presents the objective function to be maximized.
Justification
Directly from the video's blackboard.
Shown in the video
Expression
$\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$
Explanation
The video then directly presents the first-order conditions.
Justification
The video displays the resulting expression but omits the differentiation process.
Shown in the video
Expression
$H V(w^*)=-\lambda\Sigma<0$
Explanation
Finally, the video gives the second-order conditions and notes that they are used to confirm a maximum.
Justification
The video shows the result formula and annotations, but does not provide a complete statement of the criterion theorem.
Shown in the video

Conclusion

The segment focuses primarily on displaying the concluding expression, without providing the full derivation chain from the objective function to the optimality criteria.

Algebraic rewriting of the portfolio return expression

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen writes in one line: $μ_p = w^T μ + r_f(1 - w^T \cdot 1) = r_f + w^T(μ - r_f 1)$ .
Audio
Observation
The audio says: Then we rearrange it to become $r_f$ plus w times μ minus the risk-free rate times 1.

Proof

Steps

Expression
$\mu_p = w^T\mu + r_f(1 - w^T\mathbf{1})$
Explanation
Starting from the original portfolio return expression given in the video.
Justification
The Portfolio Return formula directly shown on screen.
Shown in the video
Expression
$= r_f + w^T(\mu - r_f\mathbf{1})$
Explanation
Expand and regroup $r_f(1 - w^T\mathbf{1})$ to obtain a risk-free benchmark plus an excess return term.
Justification
Algebraic expansion and combining like terms.
Derived from the video

Conclusion

Portfolio return can be written as $r_f$ plus a weighted term of the excess return vector of risky assets.

First-order condition for the mean-variance objective.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays ∂V/∂w(w^*) = ( $μ - r_f 1$ ) - λΣw^*.
Audio
Observation
The audio states that we differentiate it using the partial derivative of V with respect to w.

Uncertainties

The video does not show step-by-step intermediate steps for differentiating the quadratic form $w^TΣw$ .

Proof

Steps

Expression
$V(w) = \left[r_f + w^T(\mu - r_f\mathbf{1})\right] - \frac{\lambda}{2}w^T\Sigma w$
Explanation
Start from the objective function to be maximized.
Justification
The expression for max $V(w)$ shown on screen.
Shown in the video
Expression
$\frac{\partial}{\partial w}w^T(\mu - r_f\mathbf{1}) = \mu - r_f\mathbf{1}$
Explanation
Differentiating the linear term with respect to w yields the excess return vector.
Justification
Vector differentiation rule; though not expanded in the video, it can be verified from the on-screen formula.
Derived from the video
Expression
$\frac{\partial}{\partial w}\frac{\lambda}{2}w^T\Sigma w = \lambda\Sigma w$
Explanation
Differentiating the quadratic form yields λΣw.
Justification
Standard matrix differentiation rules; the video skips intermediate steps.
Derived from the video
Expression
$\Delta V(w^*) = \frac{\partial V}{\partial w}(w^*) = (\mu - r_f\mathbf{1}) - \lambda\Sigma w^*$
Explanation
Combine the derivatives of the two terms and set them to the first-order condition at the optimal weights.
Justification
The first-order condition shown directly on screen.
Shown in the video

Conclusion

The optimal weights w^* satisfy the first-order condition ( $μ - r_f\cdot 1$ ) - λΣw^* = 0.

Use the Hessian to determine the type of extremum.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays HV(w^*) = -λ $Σ < 0$ , and notes that this confirms we found a maximum.
Audio
Observation
The audio states that after computing the second derivative, we find the parabola opens downward, so the value we found is a maximum—that is, it is less than zero.

Uncertainties

The video uses the univariate function terminology 'opens upward/downward' to intuitively explain the multivariate Hessian test.

Intuitive argument

Steps

Expression
$\frac{\partial^2 V}{\partial w\partial w^T} = -\lambda\Sigma$
Explanation
Differentiate the first derivative again to obtain the Hessian.
Justification
This follows directly from the quadratic form structure of $V(w)$ shown on screen.
Derived from the video
Expression
$HV(w^*) = -\lambda\Sigma < 0$
Explanation
The video states that the Hessian is negative.
Justification
The second-order condition written directly on screen.
Shown in the video
Expression
$\text{negative definite Hessian} \Rightarrow \text{maximum}$
Explanation
This confirms that the stationary point is a maximum, not a minimum.
Justification
Second-order sufficient condition; the video gives an intuitive explanation using the direction in which the parabola opens.
Derived from the video

Conclusion

Under the given conditions, w^* corresponds to the maximum value of the objective function.

Rewriting the portfolio optimization problem in Lagrangian form.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays $f(x) + Σ_{j=1}^m λ_j(g_j(x) - b_j)$ , and min(w): $1/2 w^TΣw$ , Subject to: $w^Tμ = m$ , $w^T1 = 1$ .
Audio
Observation
The audio says we denote the formula just shown as fx, and what follows are its restrictions—that is, its constraints.

Uncertainties

The clip ends at the point where the example is substituted; the full solution process is not shown.

Proof

Steps

Expression
$f(x) + \sum_{j=1}^{m}\lambda_j(g_j(x) - b_j)$
Explanation
First, present the general template of the Lagrangian function.
Justification
The Lagrangian function displayed directly on screen.
Shown in the video
Expression
$\min(w): \frac{1}{2}w^T\Sigma w$
Explanation
Taking semivariance risk as the objective in the portfolio problem.
Justification
The objective given in the Constraints section on screen.
Shown in the video
Expression
$w^T\mu = m,\quad w^T\mathbf{1} = 1$
Explanation
Using target return and sum of weights as two constraint conditions.
Justification
The subject region is given directly on screen.
Shown in the video

Conclusion

This segment formulates the risk-minimizing portfolio problem within a Lagrangian optimization framework with two equality constraints.

Deriving the first-order conditions from the Lagrangian function.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen transitions from 〔 $L(w,λ,γ)$ 〕 to First-order Condition.
Audio
Observation
The presenter first explains the three terms, then states that we take the partial derivative with respect to 〔w〕.

Proof

Steps

Expression
$L(w,\lambda,\gamma)=\frac{1}{2}w^{T}\Sigma w+\lambda(m-w^{T}\mu)+\gamma(1-w^{T}\mathbf{1})$
Explanation
Starting from the Lagrangian function given in the video.
Justification
The on-screen formula is given directly.
Shown in the video
Expression
$\frac{\partial L(w,\lambda,\gamma)}{\partial w}=\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$
Explanation
Take the partial derivative with respect to w and set it equal to zero.
Justification
The instructor states, "Here we take the partial derivative with respect to w," and the screen simultaneously displays this expression.
Shown in the video

Conclusion

The first-order condition that the optimal weight must satisfy is $Σw + λμ + γ1 = 0$ .

Solve for w* from the first-order condition.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen transitions from $Σw + λμ + γ1 = 0$ to w* = $Σ^{-1}(λμ + γ1)$ .
Audio
Observation
The instructor says, "Let us rearrange this formula to isolate w."

Proof

Steps

Expression
$\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$
Explanation
Start from the first-order condition.
Justification
Obtained in the previous step.
Shown in the video
Expression
$\Sigma w=-\lambda\mu-\gamma\mathbf{1}$
Explanation
Keep terms containing w on the left-hand side and move all other terms to the right-hand side.
Justification
Algebraic rearrangement.
Derived from the video
Expression
$w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$
Explanation
Left-multiply by $Σ^{-1}$ and absorb the negative sign to obtain an explicit expression for w*.
Justification
The screen directly displays this result, and the presenter says, "Solve for w."
Shown in the video

Conclusion

The optimal weight vector can be written as w^* = $Σ^{-1}(λμ + γ1)$ .

Substitute w* into the two constraints.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen shows $μ^TΣ^{-1}(λμ + γ1)$ = ... = m and $1^TΣ^{-1}(λμ + γ1)$ = ... = 1.
Audio
Observation
The presenter says to substitute this formula back into its constraint.

Proof

Steps

Expression
$w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$
Explanation
Use the already-derived expression for the weights.
Justification
Conclusion from the previous step.
Shown in the video
Expression
$\mu^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=\lambda\mu^{T}\Sigma^{-1}\mu+\gamma\mu^{T}\Sigma^{-1}\mathbf{1}=m$
Explanation
Substitute w* into the return constraint $w^Tμ = m$ .
Justification
The on-screen formula matches the presenter's explanation.
Shown in the video
Expression
$\mathbf{1}^{T}\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})=\lambda\mathbf{1}^{T}\Sigma^{-1}\mu+\gamma\mathbf{1}^{T}\Sigma^{-1}\mathbf{1}=1$
Explanation
Substitute w* into the weight-sum constraint $w^T1 = 1$ .
Justification
The on-screen formula matches the presenter's explanation.
Shown in the video

Conclusion

Obtain two linear equations in λ and γ.

Introduce A, B, and C and solve for the multipliers.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen first defines A, B, and C, then writes λ = (Am − B) / (AC − $B^2$ ) and γ = (C − Bm) / (AC − $B^2$ ).
Audio
Observation
The instructor says that, to simplify solving this formula, the complicated parts are represented by A, B, and C, and λ and γ are then solved for separately.

Uncertainties

The video does not step-by-step demonstrate the determinant method or elimination process from the system of two equations to the formulas for λ and γ; it only gives the definitions and final results.

Proof

Steps

Expression
$A=\mathbf{1}^{T}\Sigma^{-1}\mathbf{1},\quad B=\mathbf{1}^{T}\Sigma^{-1}\mu,\quad C=\mu^{T}\Sigma^{-1}\mu$
Explanation
The three fixed scalars appearing after substituting the constraint are denoted A, B, and C respectively.
Justification
The definitions appear directly on screen.
Shown in the video
Expression
$\begin{cases}\lambda C+\gamma B=m\\\lambda B+\gamma A=1\end{cases}$
Explanation
Rewrite the previous two constraint equations using A, B, and C.
Justification
Obtained by rearranging the substitution result from the previous step.
Derived from the video
Expression
$\lambda=\frac{Am-B}{AC-B^{2}},\qquad \gamma=\frac{C-Bm}{AC-B^{2}}$
Explanation
Solve this system of two linear equations to obtain closed-form solutions for the multipliers.
Justification
The screen displays the result directly; the instructor says, 'Finally, we solve for λ and γ separately.'
Shown in the video

Conclusion

Obtain explicit expressions for λ and γ, preparing for back-substitution to find w*.

Back-substitution yields the final combined weights.

Clear evidence

Shown in the video

Evidence

Formula
Observation
At the bottom of the screen, the final expression w* = $Σ^{-1}(λμ + γ1)$ is written.
Audio
Observation
The instructor says to finally substitute lambda and gamma back into the formula for w.

Proof

Steps

Expression
$\lambda=\frac{Am-B}{AC-B^{2}},\qquad \gamma=\frac{C-Bm}{AC-B^{2}}$
Explanation
First, obtain the two multipliers.
Justification
Conclusion from the previous step.
Shown in the video
Expression
$w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$
Explanation
Substitute λ and γ back into the weight expression.
Justification
The formula at the bottom of the screen matches the instructor's explanation.
Shown in the video

Conclusion

The final optimal portfolio weights are given by w^*= $Σ^{-1}(λμ+γ1)$ .

the derivation chain from the Lagrangian function to the optimal weights

Clear evidence

Shown in the video

Evidence

Formula
Observation
The left side sequentially shows L(w, lambda, gamma), the first-order conditions, w^* = Sigma^{-1}(lambda mu + gamma mathbf{1}), the system of equations after substituting the constraints, the solutions for lambda and gamma, and finally back-substituting to obtain w^*
Audio
Observation
The presenter explains the solution procedure in the order A, B, C; lambda, gamma; weights

Proof

Steps

Expression
$L(w,\lambda,\gamma)=\frac12 w^T\Sigma w+\lambda(m-w^T\mu)+\gamma(1-w^T\mathbf{1})$
Explanation
First write the Lagrangian function with the return constraint and the budget constraint.
Justification
Given directly in the formula area on the left side of the video.
Shown in the video
Expression
$\frac{\partial}{\partial w}L(w,\lambda,\gamma)=\Sigma w+\lambda\mu+\gamma\mathbf{1}=0$
Explanation
Take the first-order partial derivative with respect to w and set it to zero to obtain the optimality condition.
Justification
Given directly as 'First order condition' on the left side of the video.
Shown in the video
Expression
$w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$
Explanation
Rearrange the first-order condition and left-multiply by Sigma^{-1} to obtain an explicit expression for the weights.
Justification
Given directly in the 'conversion' row on the left side of the video.
Shown in the video
Expression
$\begin{cases} C\lambda+B\gamma=m \\ B\lambda+A\gamma=1 \end{cases}$
Explanation
Substitute the weight expression back into the two constraints and rearrange them into a system of two linear equations in $\lambda$ and $\gamma$ .
Justification
The result appears directly in the 'After obtaining' area on the left side of the video.
Shown in the video
Expression
$\lambda=\frac{Am-B}{AC-B^2},\quad \gamma=\frac{C-Bm}{AC-B^2}$
Explanation
Solve the above system of two linear equations to obtain closed-form solutions for the two multipliers.
Justification
The formula appears directly in the lower part of the left side of the video.
Shown in the video
Expression
$w^*=\Sigma^{-1}(\lambda\mu+\gamma\mathbf{1})$
Explanation
Finally, substitute the obtained $\lambda$ and $\gamma$ back into the weight formula to get the optimal portfolio.
Justification
The 'Finally substitute back' line at the bottom of the video matches the speaker's verbal explanation.
Shown in the video

Conclusion

The video fully demonstrates the derivation path starting from the Lagrangian function, proceeding through first-order conditions, substitution of constraints, solving for the multipliers, and ultimately arriving at the optimal weights.

Worked examples · 4

Mean-variance optimal portfolio example

Clear evidence

Shown in the video

Evidence

Formula
Observation
The full-page title reads Mean Variance Optimization / Mean-Variance Optimal Portfolio, and lists Portfolio Return, Objective Function, First-Order Conditions, and Second-Order Conditions in sequence.
Audio
Observation
The audio explains risk premium, penalty term, differentiation, and second-derivative verification, segment by segment.

Problem

Construct and solve the mean-variance optimization problem in a market with N risky assets and one risk-free asset.

Given

Portfolio return formula $μ_p = w^Tμ + r_f(1 - w^T\cdot 1)$
Risk premium $μ - r_f\cdot 1$
Objective function max $V(w) = [r_f + w^T(μ - r_f 1)] - λ/2 w^TΣw$

Goal

Find the weight vector w^* that maximizes $V(w)$ , and verify that it indeed yields a maximum.

Steps

Expression
$\mu_p = r_f + w^T(\mu - r_f\mathbf{1})$
Explanation
First rewrite portfolio return as the sum of the risk-free return and an excess return term.
Justification
Algebraic identity transformation shown on screen.
Shown in the video
Expression
$\max V(w) = \left[r_f + w^T(\mu - r_f\mathbf{1})\right] - \frac{\lambda}{2}w^T\Sigma w$
Explanation
Combine profit maximization and variance penalty into a single objective function.
Justification
The optimization objective directly shown on screen.
Shown in the video
Expression
$\Delta V(w^*) = \frac{\partial V}{\partial w}(w^*) = (\mu - r_f\mathbf{1}) - \lambda\Sigma w^*$
Explanation
Take the partial derivative with respect to w and set it to zero, yielding the first-order condition.
Justification
On-screen text: First order condition.
Shown in the video
Expression
$HV(w^*) = -\lambda\Sigma < 0$
Explanation
Compute the second derivative and use the sign of the Hessian to confirm the type of extremum.
Justification
On-screen text: Second order condition (Hessian).
Shown in the video

Answer

w^* satisfies ( $μ - r_f\cdot 1$ ) - λΣw^* = 0, and corresponds to a maximum when HV(w^*) = -λ $Σ < 0$ .

Verification

The video verifies this using the second-order condition HV(w^*) = -λ $Σ < 0$ and the statement 'confirms we found Maximum'.

An example of portfolio optimization modeling under the Lagrange multiplier method

Clear evidence

Shown in the video

Evidence

Formula
Observation
The on-screen title reads 'Lagrange Method / Lagrange Multiplier Method', showing both the general form and an example of portfolio optimization constraints.
Audio
Observation
The audio states that next we examine the Lagrange multiplier method and explains that we minimize risk while requiring the portfolio return to equal m.

Uncertainties

This segment only presents modeling and does not show solving for the multipliers or the final expression for the weights.

Problem

Use the method of Lagrange multipliers to formulate the portfolio optimization problem that minimizes risk subject to return and weight constraints.

Given

General Lagrangian function: $f(x) + Σ_{j=1}^m λ_j(g_j(x) - b_j)$
Objective: min(w): $1/2 w^TΣw$
Constraints: $w^Tμ = m$ and $w^T1 = 1$

Goal

Write the portfolio optimization problem in constrained Lagrangian form.

Steps

Expression
$f(x) + \sum_{j=1}^{m}\lambda_j(g_j(x) - b_j)$
Explanation
First present the general Lagrangian function with m constraints.
Justification
Displayed directly on screen.
Shown in the video
Expression
$\min(w): \frac{1}{2}w^T\Sigma w$
Explanation
Set the objective function to be semivariance risk.
Justification
On-screen region labeled "Constraints".
Shown in the video
Expression
$w^T\mu = m,\quad w^T\mathbf{1} = 1$
Explanation
Add the target return constraint and the weight-sum constraint.
Justification
On-screen region labeled "Subject to".
Shown in the video

Answer

The Lagrangian formulation of this problem is: minimize $1/2 w^TΣw$ subject to $w^Tμ = m$ and $w^T1 = 1$ .

Verification

The video does not continue solving in this segment, so we can only verify that the modeling matches the visuals.

Implementing the Lagrange multiplier method to find optimal weights using Python

Clear evidence

Shown in the video

Evidence

Formula
Observation
The code area on the right shows the computation of $m=0.1$ , u, sigma, cor, SRS, ones, A, B, C, Lam, Gamma, and weights.
Audio
Observation
The instructor says, 'Next, let us examine the code; we should start reading it from the middle—we need to compute A, B, and C.'

Uncertainties

The precise meaning of sigma in the code is not fully clear from a single frame; it could be a vector of standard deviations or might be directly included as part of the matrix input.
The video does not display the runtime output, so the numerical values of the final weights cannot be verified.

Problem

Given the target return $m = 0.1$ , the expected return vector u, covariance-related information sigma and cor, and the all-1 vector ones, compute A, B, C, Lam, Gamma, and weights according to the earlier derivation.

Given

m equals zero point one
u equals a four-by-one array whose elements are zero point zero five, zero point zero seven, zero point one five, and zero point two seven, in order
sigma equals a four-by-one array whose elements are zero point zero seven, zero point one two, zero point three zero, and zero point six zero, in order
cor is a 4-by-4 symmetric matrix
ones equals a four-by-one array whose four elements are all one
SRS equals sigma multiplied by the correlation matrix multiplied by the transpose of sigma

Goal

Implement in code the analytical derivation of A, B, C, λ, γ, and the optimal weights w.

Steps

Expression
$SRS = sigma * correlation * sigma.T$
Explanation
First construct the covariance matrix SRS from sigma and the correlation matrix.
Justification
Show the code line directly.
Shown in the video
Expression
$A = float(np.dot(np.dot(ones.T, np.matrix(SRS).I), ones))$
Explanation
Compute the numerical version of $A = 1^T Σ^{-1} 1$ .
Justification
The code corresponds to the formula on the left, and the presenter says we need to compute A, B, and C.
Shown in the video
Expression
$B = float(np.dot(np.dot(u.T, np.matrix(SRS).I), ones))$
Explanation
Compute the numerical version of $B = μ^T Σ^{-1} 1$ .
Justification
The code line is shown directly.
Shown in the video
Expression
$C = float(np.dot(np.dot(u.T, np.matrix(SRS).I), u))$
Explanation
Compute the numerical version of $C = μ^T Σ^{-1} μ$ .
Justification
The code line is shown directly.
Shown in the video
Expression
$Lam = (A*m-B)/(A*C-np.power(B,2))$
Explanation
Compute λ using the closed-form formula.
Justification
Corresponding to the left-hand side: $λ = (A m - B) / (A C - B^2)$ .
Shown in the video
Expression
$Gamma = (C-B*m)/(A*C-np.square(B))$
Explanation
Compute γ using the closed-form formula.
Justification
Corresponding to the left-hand side: $γ = (C - B m) / (A C - B^2)$ .
Shown in the video
Expression
$weights = np.matrix(SRS).I * (Lam*u+Gamma*ones)$
Explanation
Finally, substitute back to obtain the optimal weight vector.
Justification
Corresponding to the left-hand side: w* = $Σ^{-1} (λ μ + γ 1)$ .
Shown in the video

Answer

The video shows the complete code implementation process but does not provide the numerical results after execution.

Verification

You can verify whether the implementation matches the analytical derivation by comparing SRS, u, and ones in the code term-by-term with the left-hand-side formulas.

A numerical implementation example of mean-variance optimization for four assets

Clear evidence

Shown in the video

Evidence

Formula
Observation
The code on the right gives $m=0.1$ and numerical arrays for U, sigma, and icor, and computes SRS, A, B, C, Lam, Gamma, and weights based on them.
Audio
Observation
The speaker explains the computation order for SRS, B, C, lambda, gamma, and weights.

Uncertainties

The final numerical values of weights are not displayed on screen; only the computation code is shown.
The sigma array is incompletely displayed on screen; its full dimension must be verified by reviewing the original footage.

Problem

Given target return $m = 0.1$ , expected return vector U, standard deviation vector sigma, and correlation coefficient matrix icor, find the optimal portfolio weights satisfying the budget constraint.

Given

m equals zero point one.
U is a column vector whose elements are zero point zero five, zero point zero seven, zero point one five, and zero point two seven, in that order.
sigma is the vector of standard deviations for each asset (the display is incomplete).
icor is a four-by-four correlation coefficient matrix: the first row is 1, 0.8, 0.5, 0.4; the second row is 0.8, 1, 0.7, 0.5; the third row is 0.5, 0.7, 1, 0.8; the fourth row is 0.4, 0.5, 0.8, 1.
ones is a column vector whose four elements are all 1.

Goal

Compute SRS, A, B, C, Lam, Gamma, and finally obtain weights.

Steps

Expression
$SRS = sigma * correlation * sigma.T$
Explanation
First construct the covariance matrix from standard deviations and correlation coefficients.
Justification
The code on the right matches the speaker's verbal description.
Shown in the video
Expression
$A = float(np.dot(np.dot(ones.T, np.matrix(SRS).I), ones))$
Explanation
Compute $A = \mathbf{1}^T \Sigma^{-1} \mathbf{1}$ .
Justification
The code on the right is given directly.
Shown in the video
Expression
$B = float(np.dot(np.dot(U.T, np.matrix(SRS).I), ones))$
Explanation
Compute $B = \mu^T \Sigma^{-1} \mathbf{1}$ .
Justification
The code on the right gives it directly.
Shown in the video
Expression
$C = float(np.dot(np.dot(U.T, np.matrix(SRS).I), U))$
Explanation
Compute $C = \mu^T\Sigma^{-1}\mu$ .
Justification
The code on the right gives it directly.
Shown in the video
Expression
$Lam = (A * m - B) / (A * C - np.power(B, 2))$
Explanation
Compute $\lambda$ using the closed-form formula.
Justification
The code on the right matches the formula on the left.
Shown in the video
Expression
$Gamma = (C - B * m) / (A * C - np.square(B))$
Explanation
Compute $\gamma$ using the closed-form formula.
Justification
The code on the right matches the formula on the left.
Shown in the video
Expression
$weights = np.matrix(SRS).I * ((Lam * U + Gamma * ones))$
Explanation
Substitute $\lambda$ and $\gamma$ back to obtain the optimal weight vector.
Justification
The code on the right matches the bottom formula w^* = $\Sigma^{-1}(\lambda\mu + \gamma\mathbf{1})$ .
Shown in the video

Answer

The video presents the complete numerical solution process but does not show the specific numeric values of the final weights.

Verification

You can verify process consistency by checking whether the code implements the left-side formulas—SRS, $A/B/C$ , $\lambda/\gamma$ , and w^*—in order; the final numeric results must be viewed in the original video's runtime output.

Visual events · 12

The opening title page features both the code and the geometric animation.

Clear evidence

Shown in the video

Evidence

Animation
Observation
On the title page, Python code gradually appears in the background; visible fragments include `cor = np.array([[1, 0.8, 0.5, 0.4], ...])` and `def lagrangian(u, sigma, cor, m):`.
Animation
Observation
A rotating polygon/geometric figure appears in the lower-left region, with a yellow hexagon at its center.

Uncertainties

The latter part of the code is obscured or truncated and cannot be fully read.
The mathematical meaning of the geometric figure is not explained within this segment.

Objects

Title text: Mean Variant - Lagrange / Lagrangian Method
Chinese subtitle: Mean-Variance Optimal Portfolio / Lagrange Multiplier Method
Python code snippet
Rotating polygon and central yellow hexagon

Changes

Code fades in from invisible to semi-transparent
Geometric figure rotates continuously

Invariants

Main title and subtitle positions remain largely unchanged
Black background with white/orange text color scheme remains unchanged

Interpretation

This segment mainly serves as visual packaging for the course introduction. The code contains symbols such as `lagrangian`, `sigma`, `cor`, and `m`, hinting at upcoming topics including the Lagrange method and matrix operations, but no mathematical explanation of this code appears within this segment.

Formula panel layout

Clear evidence

Shown in the video

Evidence

Diagram
Observation
The scene switches to a blackboard-style formula panel with black background, displaying formulas top to bottom: Portfolio Return, Risk Premia, objective function, First order condition, Second order condition (Hessian)
Animation
Observation
A small rotating geometric figure persists in the upper right corner

Uncertainties

No textual explanation links the small figure in the upper right corner to the formulas

Objects

Mean Variance Optimization Title
Portfolio Return Formula
Risk Premia Expression
Maximization Objective Function
First-Order Condition
Second-Order Condition
Small graphic rotating in the upper right corner

Changes

Content transitions from title page to formula board
Formulas are displayed statically, top to bottom within each panel

Invariants

Black background with white or orange text layout remains unchanged
Small graphic in the upper right corner rotates continuously

Interpretation

Visually place portfolio return, risk premia, objective function, and first- and second-order conditions on the same page, forming a continuous presentation from definitions to optimality conditions.

Switch from the mean-variance page to the Lagrange page

Clear evidence

Shown in the video

Evidence

Animation
Observation
The screen switches from the Mean Variance Optimization page to the Lagrange Method page.
Audio
Observation
The audio synchronously states, "Next, let's look at the Lagrange multiplier method."

Objects

Title: Mean Variance Optimization / Mean-Variance Optimal Portfolio
Title: Lagrange Method / Lagrange Multiplier Method
Formula block

Changes

Full-page replacement
The formula changes from portfolio return and first- and second-order conditions to the Lagrangian function and constraint examples

Invariants

The layout with black background, white text, and orange title remains unchanged
The circular decorative icon on the left remains present throughout

Interpretation

The visual transition corresponds to the lecture topic shifting from unconstrained mean-variance optimization to constrained optimization methods.

Hierarchical layout of the mean-variance formula

Clear evidence

Shown in the video

Evidence

Diagram
Observation
Portfolio Return, In a market with N risky assets and a risk-free asset, First order condition, Second order condition (Hessian), displayed from top to bottom.

Objects

Portfolio Return block
Objective function block
First order condition block
Second order condition block

Changes

Explanation proceeds vertically down the page

Invariants

All formulas are simultaneously visible on a single static page

Interpretation

The layout organizes definitions, objectives, solution conditions, and extremum tests in logical hierarchy, facilitating step-by-step derivation.

Chinese annotations on the Lagrange page

Clear evidence

Shown in the video

Evidence

Diagram
Observation
The screen labels the summation term as (Penalty), labels $1/2 w^TΣw$ as "risk", labels $w^Tμ = m$ as "m is the portfolio return", and labels $w^T1 = 1$ as "sum of all weights".

Objects

(Penalty) label
Risk annotation
m is the portfolio return annotation
Sum of all weights annotation

Changes

Abstract symbols are assigned financial meanings term by term

Invariants

The formula itself remains unchanged

Interpretation

Annotations map general optimization notation back to the portfolio optimization context, explaining that the multiplier term is a penalty, the objective term is risk, and the two constraints are respectively the return target and the budget normalization.

Switch from the problem statement to the full derivation page.

Clear evidence

Shown in the video

Evidence

Animation
Observation
First display three sections: Lagrangian Function, Constrain, and Subject to; then switch to the full derivation page.

Objects

Title: Lagrange Method / Lagrange Multiplier Method
Lagrangian Function section
Constrain section
Subject to section

Changes

The screen switches from the simplified problem statement to a complete derivation page containing the first-order conditions, substitution of the constraint, definitions of A, B, and C, and the final formula.

Invariants

The topic remains the Lagrange method for mean-variance optimization.

Interpretation

Visually present the optimization problem and constraint first, then unfold the entire solution process.

Explain each of the three parts of the Lagrangian function one by one.

Clear evidence

Shown in the video

Evidence

Animation
Observation
The presenter explains the three terms of the Lagrangian function in order, while the screen keeps the full formula visible.

Objects

The formula for 〔 $L(w,λ,γ)$ 〕
risk term
return constraint term
weight sum constraint term

Changes

Attention shifts from the entire formula to its three components.

Invariants

The formula itself remains unchanged.

Interpretation

Helps viewers associate the abstract formula with the three types of information in the optimization problem.

From the first-order condition to the explicit expression for w*

Clear evidence

Shown in the video

Evidence

Animation
Observation
The screen sequentially highlights the first-order condition and w*= $Σ^{-1}(λμ+γ1)$ .

Objects

The partial derivative of L with respect to w equals zero
the expression for w*

Changes

The derivation focus shifts from differentiation to solving for w.

Invariants

λ and γ remain as unknown multipliers.

Interpretation

Demonstrate how to convert the extremum condition into a weight expression.

Substitute w* back into the constraint.

Clear evidence

Shown in the video

Evidence

Animation
Observation
The screen continues to display the two equations after substituting the constraint, with their right-hand sides equaling m and 1, respectively.

Objects

The transpose of μ multiplied by the inverse of Σ, then multiplied by the quantity λ times μ plus γ times one, equals m.
The transpose of one multiplied by the inverse of Σ, then multiplied by the quantity λ times μ plus γ times one, equals one.

Changes

Convert the expression from weight form into equations in terms of λ and γ.

Invariants

The right-hand sides remain m and 1, respectively.

Interpretation

Explain how the original constraints become a solvable system of two equations.

Introduce A, B, and C and give the multiplier solutions.

Clear evidence

Shown in the video

Evidence

Animation
Observation
The screen first gives the definitions of A, B, and C, then provides the closed-form solutions for λ and γ.

Objects

A
B
C
lambda
gamma

Changes

The complex quadratic form expression is compressed into three scalars, followed immediately by the multiplier formula.

Invariants

The derivation goal remains finding the optimal weights.

Interpretation

Visual emphasis on simplified notation and final solvability.

Code implementation is displayed side by side with the formula.

Clear evidence

Shown in the video

Evidence

Animation
Observation
A Python code panel appears on the right, while the formula derivation remains on the left.

Objects

Formula derivation on the left
Python code on the right

Changes

Content expands from purely mathematical derivation to numerical implementation.

Invariants

Both left and right sides correspond to the same set of relationships among A, B, C, λ, γ, and w*.

Interpretation

Map the analytical formula to executable code to clarify the implementation path.

The left-side theory and the right-side code are displayed synchronously.

Clear evidence

Shown in the video

Evidence

Diagram
Observation
The entire scene maintains a two-column layout: Lagrange Method derivation on the left, Python code implementation on the right.
Animation
Observation
Approximately between 00:00:00 and 00:00:29, the circular highlight box moves sequentially to the code lines SRS, $A/B/C$ , Lam/Gamma, and weights.

Objects

Left-side Lagrangian formula derivation board.
Right-side Python code editor.
Circular highlight indicator box.

Changes

The highlight box moves from the SRS line to the $A/B/C$ line, then to the Lam/Gamma line, and finally to the weights line.
At the bottom, the watermark "WeChat Official Account: Devil's Spokesperson" appears approximately after 00:00:29.

Invariants

The left-right split-screen layout remains unchanged.
The core formula on the left remains always visible.
The overall structure of the code on the right remains unchanged.

Interpretation

Through synchronized highlighting, the video maps theoretical quantities A, B, C, $\lambda$ , $\gamma$ , and w^* on the left to corresponding code variables on the right, emphasizing that this is the numerical implementation of the same derivation.

Misconceptions · 7

Statements regarding the symbol $1$ .

Clear evidence

Shown in the video

Evidence

Audio
Observation
The narrator says, "One of them is a matrix composed of ones."
Formula
Observation
In the on-screen formula, $1$ appears in both $w^{T}*1$ and $r_f1$ , serving a vector-like role.

Misconception

Directly calling $1$ in the formula a "matrix" may not fully align with its role in the visual as a full 1 vector participating in operations.

Clarification

A more cautious interpretation is that the video uses the shorthand notation $1$ to denote an object composed of 1; in this segment's formula, it is at least compatible with vector operations. Since the video makes no further distinction, the narrator's wording should not be taken as a strict definition.

Describing the multivariate Hessian criterion as the univariate direction of opening

Clear evidence

Shown in the video

Evidence

Audio
Observation
The audio explains the relationship between the second derivative and maxima or minima using the phrases 'opening downward' and 'opening upward'.
Formula
Observation
The screen writes the Hessian condition HV(w^*) = -λ $Σ < 0$ .

Misconception

It is easy to interpret the Hessian criterion for multivariable optimization as 'opening upward' or 'opening downward' for a univariate quadratic function.

Clarification

At this point, the video explains the multivariate second-order condition using intuitive univariate language; strictly speaking, one should examine the positive or negative definiteness of the Hessian, and the formal criterion given on screen is HV(w^*) = -λ $Σ < 0$ .

The spoken notation conflicts with the on-screen notation.

Approximate timing

Shown in the video

Evidence

Audio
Observation
The audio reads the variance term as w omega w.
Formula
Observation
The screen displays $w^TΣw$ .

Uncertainties

It is unclear whether this is a slip of the tongue or a temporary adoption of another notation.

Misconception

Listeners may assume that omega and Σ are two distinct matrices.

Clarification

In this segment, the audio's 'w omega w' corresponds to the on-screen ' $w^TΣw$ '; both refer to the quadratic form expression for portfolio variance.

Mistakenly believing the final goal is λ or γ.

Clear evidence

Shown in the video

Evidence

Audio
Observation
The instructor specifically clarifies: if you haven't realized it yet, w is the weight of each asset in our optimal portfolio, and w is the final result we seek.

Misconception

Seeing many instances of solving for λ and γ in the derivation, one might mistakenly think the multipliers themselves are the final answers.

Clarification

The video explicitly states that the ultimate objective is the weight vector w, while λ and γ are merely intermediate multipliers that must later be substituted back into the w* formula.

Misinterpreting 1 in the formula as the scalar 1.

Clear evidence

Shown in the video

Evidence

Audio
Observation
The instructor explains that the matrix 1 is the ones matrix, i.e., four copies of 1 in a NumPy array.

Misconception

In $w^T1 = 1$ or in the code's ones, one might misread 1 as an ordinary number.

Clarification

In the video, 1 denotes the all-1 vector, used to compute the sum of weights; in the code, it corresponds to ones = np.array([[1,1,1,1]]).T.

Do not mistake the code variable U for a new mathematical object.

Clear evidence

Shown in the video

Evidence

Audio
Observation
The speaker says, "I use u instead of mu."
Formula
Observation
The left-side theory uses $\mu$ , while the right-side code uses U.

Misconception

When seeing U in the code, one might mistakenly think it differs from $\mu$ in the formula.

Clarification

The video explicitly states that U in the code is merely an alternative notation for $\mu$ , and both represent the same expected return vector.

SRS is not a new concept; it is simply the code name for the covariance matrix.

Clear evidence

Shown in the video

Evidence

Audio
Observation
The speaker says, 'This is the SRS in my code here, i.e., sigma times correlation times the transpose of sigma.'
Formula
Observation
The left-side formula uses $\Sigma$ , while the right-side code uses SRS.

Misconception

One might mistakenly treat SRS as a matrix independent of $\Sigma$ .

Clarification

The video clarifies that SRS is exactly the covariance matrix as implemented in the code, corresponding to $\Sigma$ in the left-side formula.

Concept relations · 27

Risk premium expression → Portfolio return formula

Clear evidence

Shown in the video

Evidence

Formula
Observation
Expression $\mu_{\pi}=r_f+w^{T}(\mu-r_f1)$ directly uses the Risk Premia expression $(\mu-r_f1)$

Application

Explanation

The risk premium expression is substituted directly into the simplified expression for portfolio return, becoming part of the portfolio return formula.

Portfolio return formula → Optimization objective function

Clear evidence

Shown in the video

Evidence

Formula
Observation
The first term of the objective function $V(w)$ is exactly $r_f+w^{T}(\mu-r_f1)$ , identical to the simplified portfolio return expression.

Application

Explanation

Maximizing the objective function directly reuses the portfolio return expression, then subtracts the $\frac{\lambda}{2}w^{T}\Sigma w$ term.

Optimization objective function → First-order condition

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen first displays $\max V(w)$ , then gives the expression for $\frac{\partial V}{\partial w}(w^*)$ .

Uncertainties

The video does not show the differentiation process.

Proof dependency

Explanation

The first-order condition is the resulting equation obtained by differentiating the displayed objective function; the segment illustrates dependencies but omits derivation details.

First-order condition → Second-order condition and maximum determination

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen sequentially displays the First order condition, followed by the Second order condition (Hessian).

Uncertainties

The video does not explain the logical connection between these two conditions in optimality testing.

Proof dependency

Explanation

The second-order condition appears immediately after the first-order condition, visually forming a continuous sequence of steps for optimality assessment; however, this segment does not clarify the complete decision chain.

$\lambda$ → Second-order condition and maximum determination

Clear evidence

Shown in the video

Evidence

Formula
Observation
$\lambda$ and $\Sigma$ both appear simultaneously in the objective function, the first-order condition, and the second-order condition.

Uncertainties

The video does not explain the meaning and properties of these two symbols.

Application

Explanation

The parameter $\lambda$ and the matrix symbol $\Sigma$ are directly used in the expression of the second-order condition $-\lambda\Sigma<0$ .

portfolio return with a risk-free asset → risk premium vector

Clear evidence

Shown in the video

Evidence

Formula
Observation
Risk premia ( $μ - r_f \cdot 1$ ) appear directly below the portfolio return formula.

Contains

Explanation

The rewritten portfolio return expression includes the risk premium vector $μ - r_f\cdot 1$ as one component.

risk premium vector → Mean-variance optimization objective function.

Clear evidence

Shown in the video

Evidence

Formula
Observation
The objective function max $V(w)$ directly uses $r_f + w^T(μ - r_f 1)$ .

Application

Explanation

The return term in the mean-variance objective function directly employs the risk premium expression.

Mean-variance optimization objective function. → First-order condition and meaning of optimal weights

Clear evidence

Shown in the video

Evidence

Formula
Observation
The first-order condition is obtained by differentiating the objective function $V(w)$ .
Audio
Observation
The audio states that to achieve the maximum, we must differentiate it.

Proof dependency

Explanation

Deriving the first-order condition relies on the objective function $V(w)$ given earlier.

First-order condition and meaning of optimal weights → Second-order condition determines a maximum

Clear evidence

Shown in the video

Evidence

Formula
Observation
The page first lists the First order condition, then the Second order condition (Hessian).
Audio
Observation
The audio states that we finally need to verify whether the solution found is a maximum or a minimum.

Proof dependency

Explanation

The second-order condition is used to verify whether the stationary point obtained from the first-order condition is a maximum.

Mean-variance optimization objective function. → Lagrangian function form

Clear evidence

Shown in the video

Evidence

Animation
Observation
The page switches from Mean Variance Optimization to the Lagrange Method.
Audio
Observation
The audio says: 'Then we look at the Lagrange multiplier method.'

Contrast

Explanation

The first half shows the unconstrained formulation using a penalty term for risk; the second half shifts to the Lagrange method using explicit constraints and multipliers.

Lagrangian function form → Risk minimization objective

Clear evidence

Shown in the video

Evidence

Formula
Observation
A general Lagrangian function is typically followed immediately by min(w): $1/2 w^TΣw$ .

Application

Explanation

The general Lagrangian framework is applied to the risk minimization objective in portfolio optimization.

Risk minimization objective → Portfolio return and weight-sum constraints

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen lists, below min(w): $1/2 w^TΣw$ , Subject to: $w^Tμ = m$ , $w^T1 = 1$ .

Contains

Explanation

The risk minimization problem is jointly constrained by two equality constraints defining the feasible weight set.

Find an answer · 30

What does $\mu_{\pi}$ represent in this video?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The voiceover explains 'the return of the portfolio, which here is mu pi'.
Formula
Observation
The screen displays the formula for $\mu_{\pi}$ .

Knowledge points

Portfolio return formula
$\mu_{\pi}$

Why is $w^{T}$ used in the formula instead of writing $w$ directly?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The voiceover states that $w^{T}$ is the transpose of the weights, 'so that matrix operations become more convenient for us'.

Knowledge points

$w$
$w^{T}$
Portfolio return formula

In the portfolio return formula, which part corresponds to the risky assets and which part corresponds to the risk-free asset?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The voiceover clarifies that the entire formula represents a portfolio composed of risky assets and a risk-free asset.
Formula
Observation
The portfolio's expected return equals the dot product of the weight vector and the vector of individual assets' expected returns, plus the risk-free rate multiplied by one minus the sum of the weights.

Knowledge points

Portfolio return formula
$\mu$
$r_f$
$w$

How is Risk Premia written in the video?

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays Risk Premia as $(\mu-r_f*1)$ .

Knowledge points

Risk premium expression
$(\mu-r_f*1)$

What does the maximization objective function look like in this segment?

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays $\max V(w)=[r_f+w^{T}(\mu-r_f1)]-\frac{\lambda}{2}w^{T}\Sigma w$ .

Uncertainties

The video does not explain the economic meaning of this objective function.

Knowledge points

Optimization objective function
$V(w)$
$\lambda$
$\Sigma$

How is the first-order condition expressed on screen?

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays $\Delta V(w^*)=\frac{\partial V}{\partial w}(w^*)=(\mu-r_f1)-\lambda\Sigma w^*$ .

Uncertainties

The video does not show the derivation process.

Knowledge points

First-order condition
$w^*$

How does the video use the second-order condition to justify that a maximum has been found?

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen displays $H V(w^*)=-\lambda\Sigma<0$ and notes that it confirms we found the maximum.

Uncertainties

The video does not explain the meaning of the matrix inequality.

Knowledge points

Second-order condition and maximum determination
$H V(w^*)$
The second-order condition confirms the maximum.

At the beginning, what knowledge is stated as necessary before studying this content?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The voiceover lists prerequisites in mathematics, finance, and programming.

Knowledge points

Mean-Variance Portfolio Optimization

What thematic clues about upcoming topics are revealed by the code snippet in the opening background?

Approximate timing

Shown in the video

Evidence

Animation
Observation
Python code appears in the background of the title slide, including `def lagrangian(u, sigma, cor, m):`.

Uncertainties

The code is not explained in this segment; its presence is only confirmed.

Knowledge points

The opening title page features both the code and the geometric animation.

What are the risk premia in the video, and why are they written as $μ - r_f\cdot 1$ ?

Clear evidence

Shown in the video

Evidence

Formula
Observation
The screen labels Risk Premia ( $μ - r_f \cdot 1$ ).

Knowledge points

portfolio return with a risk-free asset
risk premium vector

Why does the mean-variance objective subtract $λ/2 w^TΣw$ from the return?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The audio states that, to incorporate risk as a penalty term, we subtract lambda over 2 times w omega w.

Knowledge points

Mean-variance optimization objective function.

Why is λ divided by 2?

Clear evidence

Shown in the video

Evidence

Audio
Observation
The audio states that dividing lambda by 2 is solely for convenience when performing calculus.

Knowledge points

Mean-variance optimization objective function.
First-order condition for the mean-variance objective.

Coverage and review notes

Covered · Title slide and verbal introduction: course title, topic preview, prerequisite explanation, background code, and geometric animation; no fully legible mathematical derivation is visible.

Covered · Formula board: displays title 'Mean-Variance Optimization', portfolio return, risk premium, objective function, first-order conditions, and second-order conditions; the audio mainly explains the meaning of each term in portfolio return, while subsequent formulas appear visually with minimal verbal elaboration.

Covered · Explain the portfolio return formula and its reformulation, and define the risk premium.

Covered · Present the mean-variance objective function and explain λ and the variance penalty term.

Covered · Take the partial derivative of the objective function to obtain the first-order condition and explain the meaning of w^*.

Covered · Use the Hessian's second-order condition to verify that the stationary point is a maximum.

Covered · Switch to the method of Lagrange multipliers, presenting its general form and examples of portfolio optimization constraints.

Covered · The screen displays the optimization objective and two constraints; the audio states that the weights sum to 1.

Covered · The full Lagrangian function appears, and the presenter explains each of its three terms.

Covered · Take the partial derivative with respect to w to obtain the first-order condition.

Covered · Isolate w to obtain the expression for w*.

Covered · Substitute w* back into the two constraints to obtain equations in λ and γ.

Covered · Introduce A, B, and C, and give closed-form solutions for λ and γ.

Covered · Explain that the final goal is the weight vector w, and display the Python code implementation side by side.

Covered · The speaker explains that the SRS in the code is constructed from sigma, correlation, and sigma.T, corresponding to the covariance matrix in the left-hand formula.

Covered · The screen displays the Lagrangian function, first-order conditions, and the weight expression; the speaker clarifies that U in the code replaces $\mu$ in the formula.

Covered · The speaker explains that A, B, and C are computed similarly, then substituted into lambda and gamma; the screen simultaneously shows the theoretical system of equations and the code computation.

Covered · The speaker explains that lambda and gamma are finally substituted into weights to obtain the optimal portfolio; the screen displays the closed-form solution and the back-substitution formula.

Covered · The speaker notes that the derivation and code have been posted to the official WeChat account; the screen remains split left-right with a watermark at the bottom; no new mathematical steps appear in this segment.

Explore the knowledge in this video

Reviewed subject paths

Time value of money, Compound interest, Portfolio risk｜Devils-Advocate

The explanation, unpacked.

Chapters

Learning script

Knowledge cards

Course topic and prerequisites

Definition of portfolio return μπ\mu_{\pi}μπ​

Expression for Risk Premia

Display of the objective function to maximize

First-order and second-order conditions

Portfolio return with a risk-free asset

Risk premium vector

Mean-variance optimization objective function

First-order condition

Second-order condition and maximum verification

General form of the Lagrangian function

Risk minimization and portfolio constraints

The original mean-variance optimization problem

The three terms of the Lagrangian function

First-order condition with respect to weights

Explicit expression for the optimal weights

Substituting w* back into the constraints

Three intermediate quantities: A, B, and C

Closed-form solution for the Lagrange multipliers

Final formula for the optimal weights

Correspondence between the formula and Python code

Setting Up the Lagrangian Function

First-Order Conditions and the Weight Expression

Substituting into Constraints to Obtain a Two-Variable System

Detailed learning notes

μπ\mu_{\pi}μπ​

www

wTw^{T}wT

μ\muμ

rfr_frf​

111

(μ−rf∗1)(\mu-r_f*1)(μ−rf​∗1)

V(w)V(w)V(w)

λ\lambdaλ

Σ\SigmaΣ

w∗w^*w∗

HV(w∗)H V(w^*)HV(w∗)

Mean-Variance Portfolio Optimization

Portfolio return formula

Risk premium expression

Optimization objective function

First-order condition

Second-order condition and maximum determination

portfolio return with a risk-free asset

risk premium vector

Mean-variance optimization objective function.

First-order condition and meaning of optimal weights

Second-order condition determines a maximum

Lagrangian function form

first-order stationary point condition

The second-order condition confirms the maximum.

Negative definite Hessian confirms a maximum.

Oral connection between risk premium and CAPM

Meaning of the two constraints

What do the three terms of the Lagrangian function represent?

The purpose of introducing A, B, and C

The ultimate quantity to solve for.

Once the multipliers are found, the optimal portfolio weights can be recovered by substitution.

Rewriting the portfolio return formula

Displaying the transition from the objective function to the first- and second-order conditions.

Algebraic rewriting of the portfolio return expression

First-order condition for the mean-variance objective.

Use the Hessian to determine the type of extremum.

Rewriting the portfolio optimization problem in Lagrangian form.

Deriving the first-order conditions from the Lagrangian function.

Solve for w* from the first-order condition.

Substitute w* into the two constraints.

Introduce A, B, and C and solve for the multipliers.

Back-substitution yields the final combined weights.

the derivation chain from the Lagrangian function to the optimal weights

Mean-variance optimal portfolio example

An example of portfolio optimization modeling under the Lagrange multiplier method

Implementing the Lagrange multiplier method to find optimal weights using Python

A numerical implementation example of mean-variance optimization for four assets

The opening title page features both the code and the geometric animation.

Definition of portfolio return $\mu_{\pi}$

$\mu_{\pi}$

$w$

$w^{T}$

$\mu$

$r_f$

$1$

$(\mu-r_f*1)$

$V(w)$

$\lambda$

$\Sigma$

$w^*$

$H V(w^*)$

Statements regarding the symbol $1$ .

$\lambda$ → Second-order condition and maximum determination

What does $\mu_{\pi}$ represent in this video?

Why is $w^{T}$ used in the formula instead of writing $w$ directly?

What are the risk premia in the video, and why are they written as $μ - r_f\cdot 1$ ?

Why does the mean-variance objective subtract $λ/2 w^TΣw$ from the return?