Udacity · YouTube · 3:25
This 120-second whiteboard excerpt introduces Mean Variance Optimization (MVO) in an informal, conceptual way. The speaker first explains that Markowitz-style thinking about variance and covariance motivates blending assets whose movements offset one another, reducing volatility. He then qualifies this by saying the desired pattern is often short-term anti-correlation together with longer-term positive correlation. Around the middle of the clip, a hand-drawn risk-return scatter plot appears, with "risk" on the horizontal axis and "return" on the vertical axis, to represent a candidate group of assets. The speaker states the general tendency that higher-return assets also tend to be higher-risk. In the final section, a handwritten list titled "Inputs" appears, naming Expected return, Volatility, covariance, and target return. The clip defines expected return as a forward-looking estimate and volatility as a historical measure of asset variability, but it ends before giving formulas or a fuller treatment of covariance and target return. This introductory lecture segment explains the core components of Mean Variance Optimization (MVO) using a whiteboard diagram. It defines the key inputs: the covariance matrix (capturing asset correlations), and the target return (set between the minimum and maximum asset returns). The output is described as a set of asset weights that achieves the target return while minimizing risk. Visually, the presenter plots individual assets as green dots on a risk-return graph, draws a horizontal line for the target return, and marks the optimal portfolio as an orange dot to the left of all individual assets. The speaker emphasizes that considering covariance is essential for achieving this risk reduction below individual asset levels through diversification.
Use the learning inspector for key ideas and moments, or open the reading tabs for the complete notes.
Generated from the video's visuals and explanation; not verbatim speech.
The clip opens on a whiteboard titled "Mean Variance Optimization (MVO)" while the lecturer gives background in words rather than formulas. He says Markowitz contributed the idea of considering variance and covariance between individual stocks, and that this leads to wanting to blend assets with anti-correlation.
The first mathematical idea is an intuitive risk-reduction argument: if two assets move in opposite directions, one going up while the other goes down, their effects cancel and the combined portfolio has less volatility. This is presented verbally, with no equation shown.
The lecturer immediately qualifies the previous point. He says that overall one still wants the assets to move up together in general, so the practical pattern he describes is anti-correlation over the short term and positive correlation over the longer term.
From this discussion he introduces the named method of the lesson: mean variance optimization. He describes it as an algorithm for taking a potential set of assets and deciding how they should be blended together, explicitly mentioning covariance as one of the considerations.
At this point the board changes from title-only to a hand-drawn scatter plot. The vertical axis is labeled "return" and the horizontal axis is labeled "risk," and several green dots are added to represent a group of candidate assets.
Using that plot, the lecturer states a general risk-return relationship: higher-return assets tend also to be higher-risk, and moving down in risk tends to mean moving down in return. This is framed as a tendency, not a strict law.
He then returns to the central question raised by the plot: how to combine these assets into a good portfolio. That question motivates the next visual element, a list of optimizer inputs.
A heading "Inputs" appears on the left side of the board with four bullet items: Expected return, Volatility, covariance, and target return. The lecturer says these are the factors that go into a mean variance optimizer.
He begins defining the list item by item. For each stock, one must provide an expected return, meaning a forecast of what the asset is thought to provide in the future in terms of return.
He then names volatility and covariance as additional inputs. Volatility is explained as a historical measure of how variable each asset has been. Covariance and target return are listed on the board, but the clip ends before they receive a fuller definition.
The segment begins by defining the **covariance matrix** as a crucial input for Mean Variance Optimization (MVO). The speaker explains that this matrix captures the correlation of daily returns between every pair of assets in the portfolio. This information is vital because the optimizer uses it to determine the appropriate weighting for each stock, moving beyond simple individual asset analysis.
Next, the concept of **target return** is introduced as another key input. The presenter clarifies that a target return can be set anywhere within the range defined by the minimum and maximum expected returns of the individual assets. Any intermediate return level is achievable through 'blending,' or combining, the assets in different proportions. A horizontal dotted line is drawn on the risk-return scatter plot to visualize this specific target return constraint.
The focus then shifts to the **output** of the optimization process. The speaker states that the result is a specific set of weights—one for each asset. The objective of these weights is twofold: they must construct a portfolio that exactly meets the previously defined target return, while simultaneously minimizing the overall risk of that portfolio.
Finally, the video illustrates the power of **diversification** enabled by covariance. An orange dot is marked on the far left of the target return line, representing the optimal portfolio. The speaker emphasizes that this portfolio has a risk level lower than any of the individual assets (green dots) plotted on the graph. This significant risk reduction is only possible because the optimization considers the covariance between assets; relying solely on individual asset risks would prevent finding such an efficient blend.
The video introduces MVO as an algorithm for deciding how to blend a candidate set of assets into a portfolio. The spoken emphasis is on using covariance information, "among other things," rather than presenting a full optimization formula.
The lecturer gives an intuitive argument: if one asset moves up while another moves down, their effects offset, producing a lower-volatility portfolio. This is a verbal mechanism, not a symbolic derivation.
The clip does not treat anti-correlation as universally desirable. It says the preferred pattern is often anti-correlation in the short term and positive correlation in the longer term, because overall the assets should generally move up together.
A hand-drawn plot places assets as green points in a plane with horizontal axis "risk" and vertical axis "return." This visual sets up the portfolio-selection problem before any numerical inputs are listed.
The speaker states that higher-return stocks or assets generally tend to be higher-risk, and that rolling down in risk usually means accepting lower return. The wording makes this a tendency rather than a theorem.
The board lists four inputs: Expected return, Volatility, covariance, and target return. The clip defines expected return as a forward-looking estimate and volatility as a historical variability measure, while covariance and target return are named but not yet expanded.
A matrix input that records the correlation of daily returns between all pairs of assets. It allows the optimizer to account for how assets move together, which is essential for calculating portfolio-level risk accurately.
A user-defined expected return level for the portfolio. It must lie between the minimum and maximum returns of the available individual assets. Achieving this return involves blending assets in specific proportions.
The result of the optimization is a vector of asset weights. These weights define a portfolio that satisfies the target return constraint while possessing the absolute minimum possible risk (variance) among all portfolios with that same return.
By incorporating covariance, MVO can construct a portfolio with risk lower than that of any single constituent asset. This demonstrates the mathematical advantage of diversification, which is lost if only individual asset volatilities are considered.
Explore conditions, steps and evidence. Supplementary explanations are labeled separately from content shown in the video.
Handwritten title 'Mean Variance Optimization (MVO)' at the top of the whiteboard.
MVO
Mean Variance Optimization
Portfolio optimization method
Vertical axis of the scatter plot is labeled 'return'.
return
Expected return of an asset or portfolio
Vertical coordinate in the risk-return plot
Horizontal axis of the scatter plot is labeled 'risk'.
risk
Volatility or standard deviation of returns
Horizontal coordinate in the risk-return plot
The speaker says that from this work grew a number of algorithms, one of the key ones being mean variance optimization.
He describes it as a way of taking a potential set of assets and figuring out how they should be blended together by looking at their covariance among other things.
The title on screen reads "Mean Variance Optimization (MVO)".
The video defines Mean Variance Optimization informally as an algorithm for deciding how to blend a candidate set of assets into a portfolio. The spoken description emphasizes using covariance information, "among other things," rather than giving a full mathematical formulation.
Applies to a potential set of assets.
The stated purpose is to determine how those assets should be blended together.
Covariance is explicitly named as one input consideration; other inputs are mentioned only generally.
The speaker says that generally the higher return stocks or assets tend to also be the highest risk.
He adds that as we roll down the risk, we tend to also look at lower return.
A hand-drawn scatter plot has vertical axis labeled "return" and horizontal axis labeled "risk" with green dots placed across the plane.
The clip states a general tendency rather than a theorem: assets with higher expected return are usually also higher risk, and moving toward lower risk corresponds to lower return. The scatter plot visually supports this by placing multiple asset points in a risk-return plane.
Stated as a general tendency, not a universal law.
Refers to the plotted group of assets and to assets in general as discussed by the speaker.
A handwritten list appears under the heading "Inputs" with bullets: "Expected return", "Volatility", "covariance", "target return".
The speaker says, "Here's what goes into a mean variance optimizer. These are the factors that it considers."
He explains expected return as what we think in the future it's going to provide in terms of return.
He explains volatility as historically how volatile each asset has been.
The clip names "target return" on the board but does not define it before the segment ends.
The clip names covariance as an input but does not give its formula or matrix form in this excerpt.
The video presents four named inputs to a mean-variance optimizer: expected return, volatility, covariance, and target return. Of these, the speaker verbally defines expected return as a forward-looking return estimate and volatility as a historical measure of how variable an asset has been. Covariance and target return are listed but not fully explained within this clip.
These are the factors the optimizer considers according to the video.
Expected return is described per stock or asset.
Volatility is described per asset.
Covariance is listed as an input without further definition in this excerpt.
Target return is listed as an input without further definition in this excerpt.
Speaker states: 'Covariance is a matrix which shows between each asset and every other asset what is the correlation of daily returns.'
The word 'covariance' is listed under 'Inputs' on the whiteboard.
In Mean Variance Optimization, covariance is defined as a matrix that captures the correlation of daily returns between every pair of assets in the portfolio.
Applies to multiple assets in a portfolio
Speaker states: 'So the last and perhaps one of the most important inputs to this is the target return. We can target a return anywhere from the max return asset to the min return asset...'
The phrase 'target return' is listed under 'Inputs'. A horizontal dotted line is drawn across the scatter plot to represent this level.
The target return is a specified expected return level for the portfolio, which can be set anywhere between the minimum and maximum expected returns of the individual available assets.
Must be within the range of achievable returns via blending
Speaker states: 'So the output of an optimizer is a set of weights, one weight for each asset, that provides the target return but minimizes risk.'
Text 'Output: Asset weights for portfolio that minimize risk' is written below the Inputs list.
The output of the Mean Variance Optimization process is a specific set of weights assigned to each asset. These weights are chosen such that the resulting portfolio achieves the target return while having the lowest possible risk.
Given a valid target return and input parameters
The speaker says you can have a much lower risk portfolio if you combine assets that are anti-correlated or anti-varianced.
He explains that when one moves up, the other moves down, they cancel each other out, and you have much less volatility.
The phrase "anti-varianced" is spoken by the lecturer and immediately qualified with "if that is a word"; it is not formally defined.
According to the video, combining assets whose returns move in opposite directions can produce a much lower-risk portfolio because their movements partially cancel, reducing volatility.
The assets are anti-correlated or move oppositely over the relevant horizon.
The comparison concerns portfolio risk or volatility after combination.
Informal universal-style claim about what happens when such assets are combined; no formal quantification is given.
The speaker says that overall you want these assets to in general move up together.
He then says often what we're looking for is anti-correlation in the short term and positive correlation in a longer term.
The statement is presented as a common investment objective or heuristic, not as a proved theorem.
No precise time thresholds for "short term" or "longer term" are given.
The video states a desired pattern for asset selection: anti-correlation in the short term and positive correlation in the longer term, while overall the assets should generally move up together.
The discussion concerns a set of assets intended for portfolio construction.
The claim distinguishes short-term behavior from longer-term behavior.
Informal and qualitative; uses "often" and "in general" rather than exact quantifiers.
The speaker says that generally the higher return stocks or assets tend to also be the highest risk.
He adds that as we roll down the risk, we tend to also look at lower return.
The claim is explicitly hedged with "generally" and "tend to," so it is not stated as always true.
No formal model assumptions are provided in the clip.
The video asserts a general tendency that higher-return assets also tend to be higher-risk, and conversely lower risk corresponds to lower return.
The statement applies to the kinds of assets being discussed in the portfolio context.
General tendency over assets; not universal.
Speaker states: 'So in the end we're looking for a portfolio that's out here, in other words it's got risk that's even lower than any of the individual assets, but it meets the target return. And this is possible because we consider covariance. If we were only looking at individual risk, we wouldn't be able to find the right blend that would bring us out here.'
An orange dot is drawn on the far left of the dotted target return line, visually positioned to the left of all green dots (individual assets).
Considering the covariance matrix allows the optimizer to find a portfolio blend with a risk level strictly lower than that of any individual asset, while still meeting the target return.
Assets have non-perfect positive correlation (implied by the ability to reduce risk via blending)
Existential (there exists a portfolio)
Speaker states: 'We can target a return anywhere from the max return asset to the min return asset, and then anything else between those of course we can accomplish by blending.'
Any target return value lying between the minimum and maximum expected returns of the individual assets can be achieved through a linear combination (blending) of those assets.
Weights can sum to 1 (standard portfolio constraint, implied by 'blending')
Universal (for any value in the interval)
The speaker explains the mechanism verbally: when one moves up, the other moves down, they cancel each other out, and you have much less volatility.
This is an intuitive verbal derivation, not a symbolic proof.
No algebraic portfolio-variance formula is shown in the clip.
Start with two assets whose movements are opposite in direction over the period considered.
This is the condition stated by the speaker for anti-correlated or "anti-varianced" assets.
When one asset moves up and the other moves down, their effects offset each other in the combined portfolio.
The speaker explicitly says they cancel each other out.
Because of this offsetting, the combined portfolio exhibits much less volatility and therefore lower risk.
This is the conclusion stated directly in the audio.
The clip gives a verbal intuition that opposite-moving assets reduce portfolio volatility through cancellation, supporting the claim that anti-correlation can lower risk.
The speaker first discusses Markowitz adding variance and covariance between individual stocks.
He then says that out of this work grew a number of algorithms, one of the key ones being mean variance optimization.
The derivation is historical and conceptual rather than mathematical.
No equations are displayed linking Markowitz theory to MVO in this excerpt.
Markowitz introduced consideration of variance and covariance between individual stocks.
This is stated at the beginning of the clip.
That perspective motivates blending assets with offsetting behavior to reduce risk.
The speaker connects anti-correlation to lower-risk portfolios in the opening explanation.
From this line of work, several algorithms emerged, including mean variance optimization.
The speaker explicitly says that out of this work grew a number of algorithms, one key one being MVO.
MVO is then framed as a method for determining how a set of assets should be blended, using covariance among other considerations.
This is the spoken definition near the end of the opening section.
The clip presents MVO as an algorithmic development arising from Markowitz-style attention to variance and covariance, with the practical goal of asset blending.
The video sequentially builds the MVO concept: listing inputs, plotting assets, drawing the target return line, defining the output, and marking the optimal portfolio point.
Identify individual assets as points in the risk-return plane (green dots).
Initial setup of the optimization universe.
Define a specific expected return level using a horizontal line (dotted line).
Setting the constraint for the optimization problem.
Search for a portfolio (blend of assets) that lies on the constraint line but has the minimum x-coordinate (risk).
Applying the objective function (minimize risk) subject to the constraint (target return).
Locate the optimal portfolio (orange dot) which is to the left of all individual asset points on the same return level.
Demonstrating the mathematical result of incorporating covariance into the variance calculation.
The visual process demonstrates that MVO finds a global minimum variance portfolio for a given return, leveraging diversification effects captured by covariance.
A hand-drawn coordinate system appears with vertical axis labeled "return" and horizontal axis labeled "risk".
Several green dots are plotted in the first quadrant-like region of the sketch.
The speaker says, "So let's consider this group of assets here and how we might combine them or allocate funds to them to provide a good portfolio."
No numerical coordinates are written for the dots.
The exact number of plotted assets is visually approximate because the sketch is hand-drawn and some points are close together.
Use a drawn set of assets to motivate how one might combine or allocate funds among them to form a good portfolio.
A group of assets is represented by green points on a hand-drawn plot.
The vertical axis is labeled "return".
The horizontal axis is labeled "risk".
Introduce a concrete visual example of candidate assets before discussing optimizer inputs.
Draw a risk-return plane with "risk" on the horizontal axis and "return" on the vertical axis.
This is the visible diagram introduced in the clip.
Place multiple green points in the plane to represent different assets.
The speaker refers to "this group of assets here" while the plotted points are on screen.
Use the picture to frame the question of how to combine or allocate funds among these assets.
This is exactly the spoken purpose attached to the plot.
Note the general tendency that higher-return assets also tend to be higher-risk, which the plot is meant to evoke.
The speaker adds this interpretation immediately after introducing the asset group.
The example is not solved numerically; it serves as a visual setup for discussing portfolio construction and the inputs needed by a mean-variance optimizer.
The clip verifies the example only conceptually through the spoken framing and the hand-drawn axes and points; no computation or explicit portfolio result is shown.
The screen shows the handwritten title "Mean Variance Optimization (MVO)" with a horizontal underline and otherwise blank white space below.
Handwritten title "Mean Variance Optimization (MVO)"
Horizontal underline beneath the title
Blank whiteboard area below
No new mathematical objects are added during most of this interval.
The visual remains static while the speaker discusses Markowitz, variance, covariance, and anti-correlation.
The topic title stays fixed on screen.
No formulas or graphs are yet displayed.
The opening visual functions as a topic label while the lecture provides conceptual background verbally.
A hand enters and draws a vertical axis, then a horizontal axis, then labels them "return" and "risk".
Green dots are added one by one to form a scattered set of asset points.
The completed sketch shows a risk-return plane populated by multiple green points.
The exact count of green dots is approximate due to hand-drawing and overlapping placement.
No numeric scale is shown on either axis.
Vertical axis labeled "return"
Horizontal axis labeled "risk"
Multiple green dots representing assets
Presenter's hand and pen
First the axes are drawn.
Next the axis labels are written.
Then green asset points are added sequentially.
Finally the hand gestures over the plotted points while the speaker discusses combining assets and the risk-return tendency.
The axes remain labeled "return" vertically and "risk" horizontally.
The plotted points stay as discrete asset markers without connecting lines or curves.
The animation turns the abstract discussion of asset blending into a two-dimensional representation where each asset has a risk level and a return level.
A new handwritten heading "Inputs" appears on the left side of the board.
Four bullet items are visible: "Expected return", "Volatility", "covariance", "target return".
The presenter's pen points toward the list and later toward the scatter plot while the speaker explains the terms.
The clip ends before any further elaboration on covariance or target return.
The handwriting uses lowercase for "covariance" and "target return" while the first two bullets are capitalized.
Heading "Inputs"
Bullet list with four items
Scatter plot still visible on the right
Presenter's pen
The left side of the board gains a structured list of inputs.
The speaker begins defining the listed terms in order.
The pen alternates between indicating the list and referencing the plotted assets.
The risk-return plot remains on screen to the right.
The four named inputs stay fixed once written.
This visual event shifts the lecture from qualitative motivation to the practical ingredients required by a mean-variance optimizer.
A 2D Cartesian coordinate system with 'risk' on the x-axis and 'return' on the y-axis. Multiple green dots are scattered in the first quadrant.
Axes (risk, return)
Green dots (individual assets)
None; static background for the explanation
Coordinate system orientation
Relative positions of green dots
Represents the feasible set of individual assets characterized by their expected return and risk.
A hand draws a horizontal dotted line across the scatter plot at a specific y-value.
Horizontal dotted line
Line appears from left to right
Y-intercept (constant return level)
Visually represents the equality constraint for the optimization problem.
An orange dot is drawn on the far left end of the dotted target return line.
Orange dot
Dot appears
Position on the target return line
Leftmost position relative to other points on that line
Represents the solution vector (portfolio weights) that minimizes variance subject to the target return constraint.
After explaining the benefit of anti-correlation, the speaker adds that overall you want these assets to in general move up together.
He then distinguishes short-term anti-correlation from longer-term positive correlation.
One might infer from the opening explanation that pure anti-correlation is always the goal in portfolio construction.
The video explicitly refines this by saying that overall the assets should generally move up together, and that the desired pattern is often anti-correlation in the short term and positive correlation in the longer term.
The speaker says generally the higher return stocks or assets tend to also be the highest risk.
The statement could be mistaken for an absolute rule that every higher-return asset must be higher-risk.
The wording in the video is hedged with "generally" and "tend to," presenting the relation as a common tendency rather than a strict law.
Speaker states: 'If we were only looking at individual risk, we wouldn't be able to find the right blend that would bring us out here.'
Believing that portfolio risk can be minimized effectively by only considering the individual risks (volatilities) of the assets.
Without accounting for the covariance (correlation) between assets, it is impossible to identify blends that reduce overall portfolio risk below the level of the safest individual asset.
The speaker begins with Markowitz's consideration of variance and covariance and later says that out of this work grew mean variance optimization.
The clip presents the conceptual origin of MVO in variance and covariance thinking before listing the concrete inputs used by the optimizer.
MVO is described as blending assets by looking at their covariance among other things.
"covariance" appears in the Inputs list.
The clip does not show the covariance formula, so the link is verbal and list-based rather than symbolic.
The definition of MVO as an asset-blending method is operationalized in the later list of inputs, where covariance is explicitly included.
The scatter plot shows assets positioned by risk and return.
The Inputs list includes Expected return and Volatility.
The speaker moves from considering the plotted group of assets to asking how to combine them and then lists optimizer inputs.
The video does not explicitly state that the horizontal axis "risk" corresponds mathematically to volatility or standard deviation; it only places those ideas in nearby discussion.
The plotted risk-return representation motivates the need for numerical inputs such as expected return and volatility in the optimizer.
The speaker notes the general higher-return/higher-risk tendency and then asks how to combine the assets into a good portfolio.
The risk-return tendency frames the optimization problem that MVO is then introduced to solve.
Speaker links the possibility of finding a lower-risk portfolio directly to considering covariance.
The existence of an optimal portfolio with risk lower than individual assets (the output) depends mathematically on the inclusion of the covariance matrix in the optimization logic.
Speaker defines the output as weights that 'provide the target return'.
The target return serves as the primary constraint applied when determining the optimal asset weights.
The speaker defines mean variance optimization as a way of blending a potential set of assets using covariance among other things.
The speaker explains that opposite movements cancel each other out and reduce volatility.
The board lists Expected return, Volatility, covariance, and target return under Inputs.
A hand-drawn plot labels the axes "return" and "risk" and shows green asset points.
The speaker says expected return is what we think in the future it's going to provide in terms of return.
The speaker says volatility is simply historically how volatile each asset has been.
Definition of covariance as a matrix showing correlations between assets.
Explanation of setting target return between min and max asset returns.
Explanation that considering covariance allows finding a blend with lower risk than individual assets.
Covered · Opening verbal discussion of Markowitz, variance, covariance, anti-correlation, and the short-term versus long-term correlation idea while only the title is on screen.
Covered · The speaker introduces mean variance optimization as an algorithm for blending assets using covariance among other things.
Covered · A risk-return scatter plot is drawn and used to introduce a group of assets and the general higher-return/higher-risk tendency.
Covered · The Inputs list appears and the speaker begins defining expected return and volatility; covariance and target return are listed but not further developed before the clip ends.
Covered · Definition of covariance matrix input.
Covered · Definition and range of target return input.
Covered · Definition of optimizer output (asset weights).
Covered · Explanation of risk reduction via covariance and visual marking of optimal portfolio.
Covered · Closing remarks transitioning to next topic; no new mathematical content.
Reviewed subject paths
Continue with related videos in the same subject.
YouTube8:13
YouTube6:03
YouTube5:16