Time Series Talk : ARIMA Model

1. Audio
   Observation

   The speaker says the number of anchors sold every month is given by a variable a sub t.

2. Diagram
   Observation

   The upper graph has vertical axis labeled $a_t$ and horizontal axis labeled t.

$a_t$

Number of anchors sold at time period t, in thousands.

t is the month or time period.

1. Audio
   Observation

   The speaker says t being the month or the time period you're in.

2. Diagram
   Observation

   Both graphs use t as the horizontal-axis label.

t

Time index, specifically month or time period.

Discrete time periods in the example.

1. Diagram
   Observation

   The lower graph has vertical axis labeled $z_t$.

2. Formula
   Observation

   The displayed relation is $z_t = a_{t+1} - a_t$.

$z_t$

Differenced series obtained from consecutive values of $a_t$.

Defined by the displayed formula $z_t = a_{t+1} - a_t$.

1. Audio
   Observation

   The speaker introduces the ARIMA model and later states that ARIMA stands for Auto Regressive Integrated Moving Average.

2. Diagram
   Observation

   The title on the board reads ARIMA MODEL.

ARIMA

Auto Regressive Integrated Moving Average model.

Time series model class discussed in the clip.

1. Audio
   Observation

   The speaker asks what the I is doing here and later says the I stands for integrated.

2. Diagram
   Observation

   The left-side acronym list shows I with the word Integrated written next to it.

I

Integrated component of ARIMA.

Part of the ARIMA acronym.

1. Audio
   Observation

   The speaker identifies AR as autoregressive.

2. Diagram
   Observation

   The left-side list pairs A with Auto and R with Regressive.

AR

Auto Regressive part of the model name.

Component of ARMA/ARIMA naming.

1. Audio
   Observation

   The speaker identifies MA as moving average.

2. Diagram
   Observation

   The left-side list pairs M with Moving and A with Average.

MA

Moving Average part of the model name.

Component of ARMA/ARIMA naming.

1. Diagram
   Observation

   A partially visible formula on the right includes $\phi_1$ and an ellipsis.

1. The full formula is cut off and not fully legible.

2. The speaker does not explain this formula in the clip.

$\phi_1$

A coefficient appearing in a partially visible ARIMA-related formula on the right side of the board.

Only partially shown; full context unavailable in this clip.

1. Diagram
   Observation

   A partially visible formula on the right includes $\varepsilon_{t-1}$ and $\varepsilon_t$.

1. The full formula is cut off and not fully legible.

2. No spoken explanation of these symbols occurs in the clip.

$\varepsilon_t$

Error or innovation term appearing in a partially visible formula on the right side of the board.

Only partially shown; full context unavailable in this clip.

1. Diagram
   Observation

   Top center of the whiteboard shows the title "ARIMA MODEL".

ARIMA

Name of the time-series model being introduced; the left column expands it as Auto Regressive Integrated Moving Average.

Time series modeling notation

1. Diagram
   Observation

   Upper right drawing is an anchor labeled "= $q_t$".

2. Audio
   Observation

   The speaker says the example concerns "the number of anchors sold in thousands".

$q_t$

Original time series value at time t, represented by the anchor drawing and used in the differencing formula.

Observed time index t; values are described verbally as numbers of anchors sold in thousands.

1. Formula
   Observation

   Between the two graphs the board shows "$z_t = q_{t+1} - q_t$".

2. Audio
   Observation

   The speaker defines a new series $z_t$ as the difference from one timestamp to the previous timestamp.

$z_t$

Differenced time series obtained by subtracting the current original value from the next original value.

Time index t; constructed from consecutive q-values.

1. Audio
   Observation

   The speaker says they will talk about a model in time series analysis called the ARIMA model.

2. Diagram
   Observation

   The board title reads ARIMA MODEL.

The clip introduces ARIMA as a time series model. The speaker notes that AR and MA are already familiar as autoregressive and moving average, and focuses on explaining the additional I component.

1. The discussion is within time series analysis.

1. Audio
   Observation

   The speaker says ARIMA stands for Auto Regressive Integrated Moving Average, and the I stands for integrated.

2. Diagram
   Observation

   The left-side list spells out Auto, Regressive, Integrated, Moving, Average.

ARIMA is defined verbally and visually as Auto Regressive Integrated Moving Average. The new element relative to ARMA is the Integrated term, represented by I.

1. ARIMA model introduction

1. Audio
   Observation

   The speaker says AR, MA, or ARMA models require the time series to be stationary.

2. Audio
   Observation

   The speaker summarizes stationarity as needing constant mean, constant variance over time, and no seasonality.

1. The clip gives a summary rather than a formal definition.

2. The speaker says 'and so on,' indicating additional conditions are omitted.

The video states that AR, MA, and ARMA models require stationarity. In the spoken summary, stationarity means the series has constant mean, constant variance over time, and no seasonality.

1. Applies to AR, MA, and ARMA models according to the speaker.

1. Audio
   Observation

   The speaker says the series seems stationary except for a moving mean, so a straight ARMA model cannot be used.

2. Audio
   Observation

   The speaker says ARIMA is used in situations where things seem stationary except for a moving mean.

The Integrated part of ARIMA is presented as the mechanism for handling series that are otherwise suitable for ARMA modeling but have a nonconstant mean trend. The clip frames ARIMA as extending ARMA to such cases.

1. Used when the series appears stationary apart from a moving mean or trend.

1. Stationarity requirement for AR/MA/ARMA
2. Meaning of the ARIMA acronym

1. Diagram
   Observation

   Left column spells out "Auto", "Regressive", "Integrated", "Moving", "Average" under the title "ARIMA MODEL".

2. Audio
   Observation

   The speaker says, "Before you freak out thinking that we're talking about integrals here, integrated in this context just means that instead of predicting the time series itself, you're going to be predicting differences of the time series from one timestamp to the previous timestamp."

In this clip, "Integrated" does not refer to calculus integration. It refers to modeling differences of the original time series rather than the raw series itself. The board visually supports this by placing the differenced series below the original series and writing the transformation between them.

1. Applies to the ARIMA terminology as explained in this introductory segment.

2. The explanation is given at a high level rather than through formal theory.

1. Formula
   Observation

   The board shows "$z_t = q_{t+1} - q_t$" between the upper and lower plots.

2. Audio
   Observation

   The speaker says, "We're gonna create a new time series called Z sub T, which is simply simply gonna be defined as A sub T plus one minus A sub T."

3. Audio
   Observation

   He then gives the example of February anchors minus January anchors to produce one point of the new series.

1. The spoken letter sounds like "A" while the written formula uses q; the visual formula is clearer and is taken as the board notation.

The video introduces a transformed series $z_t$ formed by taking successive differences of the original series. On the board this is written explicitly as $z_t = q_{t+1} - q_t$, and the speaker explains it as subtracting the previous month's value from the current month's value to obtain one point of the new series.

1. Requires an ordered time series with values indexed by time.

2. The clip uses a concrete monthly example to explain adjacent-time differencing.

1. Meaning of "Integrated" in ARIMA

1. Audio
   Observation

   The speaker says the original series "seems to be pretty much linear" with noise around a linear trend.

2. Audio
   Observation

   He continues, "to get from one point of a linear function to the next, you just add some constant value," and therefore differences should hover around a constant.

3. Diagram
   Observation

   The upper plot is drawn as an orange rising jagged line, visually consistent with a trend plus noise.

The clip motivates differencing by observing that the original series looks approximately linear apart from noise. For a linear function, moving one step forward changes the value by a constant amount, so taking successive differences should remove the trend and leave values fluctuating around a constant level.

1. The argument is presented intuitively for an approximately linear trend.

2. The video explicitly says it is not going deeply into theory here.

1. Definition of the differenced series $z_t$

1. Audio
   Observation

   The speaker says, "So we've solved the problem of our time series not being stationary. Now it has a constant mean over time."

2. Diagram
   Observation

   The lower plot is drawn as a pink jagged line oscillating around a horizontal level rather than rising.

3. Animation
   Observation

   Near the end, the presenter adds a horizontal dashed line across the lower plot to mark the level around which the differenced series hovers.

1. The clip states the conclusion verbally but does not define stationarity formally within this excerpt.

After explaining why differences of a linear trend should be roughly constant, the video concludes that the transformed series now has a constant mean over time and therefore is no longer non-stationary in the same way as the original trending series. The lower graph visually reinforces this by showing fluctuations around a stable level.

1. This is the video's high-level interpretation of the differenced plot.

2. Formal stationarity conditions are not spelled out in this clip.

1. Definition of the differenced series $z_t$
2. Why differencing helps when the original series has a linear trend

1. Audio
   Observation

   The speaker defines ARIMA as having three parameters p, d, q, where p is AR order, q is MA order, and d is integrated order.

2. Formula
   Observation

   ARIMA(1,1,1) and p,d,q are written on the board.

An ARIMA model extends ARMA by adding a differencing order d. In the notation ARIMA(p,d,q), p is the autoregressive order, d is the integration or differencing order, and q is the moving-average order. The clip uses ARIMA(1,1,1) as the simplest example.

1. p is the order of the AR part

2. q is the order of the MA part

3. d is the order of the integrated part

4. the worked example in this clip is $p=1$,$d=1$,$q=1$

1. Audio
   Observation

   The speaker says $d=1$ means a difference one because they just took the first difference.

2. Formula
   Observation

   The board shows $Z_t = q_{t+1} - q_t$.

Taking the first difference replaces the original level series $q_t$ with a new series $Z_t$ formed from consecutive differences. This is the integrated step corresponding to $d=1$.

1. The original series is indexed by time

2. one difference is taken

3. this matches $d=1$ in ARIMA(1,1,1)

1. Definition of ARIMA(p,d,q)

1. Audio
   Observation

   The speaker says you can instead want the second difference, creating a new series like $w_t = z_{t+1} - z_t$, and you can do a third difference, though usually a first one suffices depending on the task.

1. No explicit formula for third differencing is written on the board.

If first differencing is not enough, the differenced series can itself be differenced again. The clip describes a second difference as forming $W_t = Z_{t+1} - Z_t$, and notes that higher orders are possible but often unnecessary.

1. Applies after an initial differencing step

2. used when more differencing is needed

3. the speaker says usually first differencing suffices but it depends on the task

1. First differencing in ARIMA

1. Audio
   Observation

   The speaker writes and explains the simplest form of ARIMA as 1,1,1 given by $Z_t$ = phi_1 $Z_{t-1}$ + theta_1 epsilon_{t-1} + epsilon_t.

2. Formula
   Observation

   The completed equation is visible on the board.

For the example ARIMA(1,1,1), the modeled differenced series $Z_t$ equals an autoregressive term in the lagged differenced value plus a moving-average term in the lagged error plus the current error.

1. This is the specific $p=1$,$d=1$,$q=1$ case

2. $Z_t$ must already be the differenced series

3. the equation is written for the differenced variable, not the original level variable

1. Definition of ARIMA(p,d,q)
2. First differencing in ARIMA

1. Audio
   Observation

   The speaker says you can't just straight up use an AR or MA or even ARMA model here because those models all require the time series to be stationary.

2. Audio
   Observation

   The speaker says the mean is shifting upward in a linear fashion, so the straight ARMA model cannot be used.

1. The claim is stated informally rather than proved in this clip.

If a time series is not stationary, then AR, MA, and ARMA models cannot be applied directly to it.

1. The series under discussion has a nonconstant mean trend.

For the AR, MA, and ARMA models mentioned in the clip.

1. Audio
   Observation

   The speaker says you use the ARIMA model in situations like this where things seem like they're stationary except for a pesky moving mean.

1. This is a practical motivation statement, not a formal theorem.

ARIMA is appropriate in situations where a series looks stationary except for a moving mean.

1. The series otherwise satisfies the stationarity-like conditions discussed in the clip.

For situations of the type described by the speaker.

1. Audio
   Observation

   The speaker says that for a linear function, going from one point to the next adds a constant value, so the differences are always constant.

1. The statement is made informally for the plotted example rather than proved algebraically in this clip.

If a function is linear, then moving from one input to the next changes the output by a constant amount, so successive differences are constant.

1. The underlying trend is treated as linear.

2. Consecutive observations are taken at equal time steps.

For consecutive points of a linear function.

1. Audio
   Observation

   The speaker says that after taking differences, the series is expected to hover around some constant and that this solves the non-stationarity problem.

2. Diagram
   Observation

   The lower plot shows the differenced series fluctuating around a level, and a dashed horizontal line is added near the end.

1. The clip presents this as an intuitive conclusion from the example, not as a general theorem with full assumptions.

For the example shown, differencing the original trending series produces a new series that hovers around a constant mean over time.

1. The original series has an approximately linear trend with noise.

2. The transformation $z_t = q_{t+1} - q_t$ is applied.

Stated for the illustrated time series example.

1. Audio
   Observation

   The speaker says the series has the same constant standard deviation over time and no seasonality, so they are free to use the ARIMA models with more confidence.

1. The clip gives this as a spoken precondition but does not formalize stationarity mathematically.

If the series has constant standard deviation over time and no seasonality, then the speaker treats ARIMA modeling as more appropriate.

1. constant standard deviation over time

2. no seasonality

For the series under discussion in this clip.

1. Audio
   Observation

   The speaker says in this case what we did was d equals one, a difference one, because we just took the first difference.

In the displayed ARIMA(1,1,1) example, $d=1$ corresponds to taking the first difference of the original series.

1. The model is ARIMA(1,1,1)

2. the series has been differenced once

For the example shown on the board.

1. Audio
   Observation

   Speaker says, "the ARIMA model is not that much crazier than just an ARMA model."

2. Formula
   Observation

   The board keeps the same ARMA-style structure after defining $Z_t$ from $q_t$.

In this presentation, ARIMA is described as essentially an ARMA model used after transforming a trending time series by differencing.

1. The time series has an obvious linear trend upward or downward.

2. The differenced series is the object to which the ARMA equation is applied.

For the ARIMA setup shown on the board.

1. Audio
   Observation

   Speaker says the choice of differencing is basically when the time series becomes stationary.

2. Audio
   Observation

   Speaker adds that if first differences are not stationary, one could try second or third differences, but typically keep the order as low as possible.

1. The clip does not define stationarity formally or specify a statistical test.

The differencing order d should be chosen so that the transformed time series becomes stationary, preferably using the lowest such order.

1. The modeler can inspect whether the differenced plot is stationary.

2. Higher differencing orders are available if lower ones do not produce stationarity.

Practical rule stated for choosing d in ARIMA modeling.

1. Audio
   Observation

   Speaker says changing p or q would be the same thing as an ARMA model.

2. Audio
   Observation

   Speaker says more lags in the AR part correspond to p and more error lags in the MA part correspond to q.

In the ARIMA model shown, p determines the number of autoregressive lags and q determines the number of moving-average error lags, just as in ARMA.

1. The series has already been differenced according to d.

For the AR and MA components of ARIMA(p,d,q).

1. Diagram
   Observation

   The upper plot shows $a_t$ versus t with an upward-trending jagged orange line.

2. Formula
   Observation

   A downward arrow between plots is labeled $z_t = a_{t+1} - a_t$.

3. Diagram
   Observation

   The lower plot shows $z_t$ versus t with a flatter jagged purple line around a horizontal level.

1. The clip visually motivates the transformation but does not prove stationarity of $z_t$.

1. Expression

   $$a_t$$
   Explanation

   Start with the original monthly sales series plotted against time.

   Justification

   This is the variable introduced by the speaker and labeled on the upper graph.

   Shown in the video
2. Expression

   $$z_t = a_{t+1} - a_t$$
   Explanation

   Form a new series by subtracting each value from the next value.

   Justification

   This relation is explicitly written on the board between the two graphs.

   Shown in the video
3. Expression

   $$z_t$$
   Explanation

   The resulting lower plot represents the transformed series.

   Justification

   The lower graph is labeled $z_t$ and is presented as the output of the displayed subtraction rule.

   Shown in the video

The clip uses the displayed formula $z_t = a_{t+1} - a_t$ to show how a trending series $a_t$ is converted into a flatter series $z_t$, motivating the Integrated step before ARMA modeling.

1. Audio
   Observation

   The speaker moves from describing the original series as linear-with-noise to explaining constant increments of a linear function, then to the claim that differences hover around a constant.

2. Diagram
   Observation

   The board contrasts an upper rising series with a lower fluctuating series linked by $z_t = q_{t+1} - q_t$.

1. This is an intuitive derivation rather than a formal proof; no algebraic expansion of $q_{t+1}-q_t$ is shown on screen.

1. Expression

   $$$$
   Explanation

   Observe that the original plotted series appears approximately linear apart from noise.

   Justification

   The speaker explicitly describes the upper series as "pretty much linear" with noise around a linear trend.

   Shown in the video
2. Expression

   $$$$
   Explanation

   Recall that a linear function changes by a constant amount from one step to the next.

   Justification

   The speaker states that to get from one point of a linear function to the next, you add some constant value.

   Shown in the video
3. Expression

   $$z_t = q_{t+1} - q_t$$
   Explanation

   Apply successive differencing to the original series to form a new series.

   Justification

   This transformation is written on the board and described verbally as creating Z sub T from adjacent values.

   Shown in the video
4. Expression

   $$$$
   Explanation

   Conclude that the differenced values should hover around a constant level rather than keep rising.

   Justification

   The speaker says that if we take the difference between one timestamp and the previous timestamp, we expect it to hover around some constant.

   Shown in the video
5. Expression

   $$$$
   Explanation

   Interpret the lower plot as the transformed series and state that it now has a constant mean over time.

   Justification

   The speaker points to the lower plot and says, "So we've solved the problem of our time series not being stationary. Now it has a constant mean over time."

   Shown in the video

The clip's reasoning is that differencing removes the visible linear trend, leaving a series that fluctuates around a stable mean.

1. Audio
   Observation

   The speaker builds the equation term by term and identifies each part as AR, MA, or current error.

2. Formula
   Observation

   The board changes from blanks to $Z_t$ = phi_1 $Z_{t-1}$ + theta_1 epsilon_{t-1} + epsilon_t.

1. Expression

   $$$$
   Explanation

   Start from the general parameter meaning of ARIMA(p,d,q): p controls the AR order, d controls differencing, and q controls the MA order.

   Justification

   Stated directly by the speaker when defining ARIMA(1,1,1).

   Shown in the video
2. Expression

   $$Z_t = q_{t+1} - q_t$$
   Explanation

   Because $d=1$, replace the original level series by its first difference and model that new series.

   Justification

   Spoken explanation plus the formula already written on the board.

   Shown in the video
3. Expression

   $$\phi_1 Z_{t-1}$$
   Explanation

   Since $p=1$, include one autoregressive lag term in the differenced series.

   Justification

   The speaker explicitly calls this the autoregressive bit.

   Shown in the video
4. Expression

   $$\theta_1 \epsilon_{t-1}$$
   Explanation

   Since $q=1$, include one moving-average lag term in the error.

   Justification

   The speaker explicitly calls this the moving average bit.

   Shown in the video
5. Expression

   $$+ \epsilon_t$$
   Explanation

   Add the current-time error term to complete the model equation.

   Justification

   The speaker says we of course have our error in the current time period.

   Shown in the video
6. Expression

   $$Z_t = \phi_1 Z_{t-1} + \theta_1 \epsilon_{t-1} + \epsilon_t$$
   Explanation

   Combine the pieces into the full ARIMA(1,1,1) equation for the differenced series.

   Justification

   This is the final equation written on the board.

   Shown in the video

The clip derives the concrete ARIMA(1,1,1) model by applying first differencing and then adding one AR lag and one MA lag to the differenced series.

1. Formula
   Observation

   The step-by-step algebraic expansion is written on the bottom left of the board.

2. Audio
   Observation

   The speaker verbally walks through the substitution process: '$a_k$ is equal to $z_{k-1}$ plus $a_{k-1}$. We just keep going. $a_{k-1}$ is equal to $z_{k-2}$ plus $a_{k-2}$...'

1. Expression

   $$z_t = a_{t+1} - a_t$$
   Explanation

   Start with the definition of the first difference.

   Justification

   Definition of differencing shown on the board.

   Shown in the video
2. Expression

   $$a_{t+1} = z_t + a_t$$
   Explanation

   Rearrange the definition to solve for the next value of the original series.

   Justification

   Algebraic manipulation.

   Derived from the video
3. Expression

   $$a_k = z_{k-1} + a_{k-1}$$
   Explanation

   Substitute $t = k-1$ to express the target future value $a_k$.

   Justification

   Application of the rearranged formula.

   Shown in the video
4. Expression

   $$a_k = z_{k-1} + (z_{k-2} + a_{k-2})$$
   Explanation

   Substitute the expression for $a_{k-1}$ into the equation.

   Justification

   Recursive substitution.

   Shown in the video
5. Expression

   $$a_k = \sum_{i=1}^{k-l} z_{k-i} + a_l$$
   Explanation

   Continue substituting backwards until reaching the last known observation $a_l$. The sum collects all intermediate differenced terms.

   Justification

   Telescoping sum / recursive expansion.

   Shown in the video

The future value $a_k$ can be calculated as the sum of predicted differences $z_{k-1}$ down to $z_{l+1}$ plus the last observed actual value $a_l$.

1. Formula
   Observation

   The lower-left derivation starts from $q_k = Z_{k-1}+q_{k-1}$ and expands backward.

2. Audio
   Observation

   Speaker says that after doing the addition, we get the best guess for the original quantity at time k.

1. Expression

   $$Z_{k-1}=q_k-q_{k-1}$$
   Explanation

   The board's definition $Z_t=q_{t+1}-q_t$ is evaluated at $t=k-1$.

   Justification

   Substitution into the displayed first-difference definition.

   Derived from the video
2. Expression

   $$q_k=Z_{k-1}+q_{k-1}$$
   Explanation

   Rearrange the previous equation to solve for $q_k$.

   Justification

   Algebraic rearrangement.

   Shown in the video
3. Expression

   $$q_k=Z_{k-1}+Z_{k-2}+q_{k-2}$$
   Explanation

   Apply the same relation to $q_{k-1}$.

   Justification

   Repeated substitution using the first-difference identity.

   Shown in the video
4. Expression

   $$q_k=\sum_{i=1}^{k-1}Z_{k-i}+q_1$$
   Explanation

   Continue substituting backward until reaching the initial value $q_1$.

   Justification

   Telescoping accumulation of first differences.

   Shown in the video

The original level $q_k$ is recovered by adding the initial value $q_1$ to the accumulated prior differenced values.

1. Audio
   Observation

   The speaker sets up a salesman of anchors who sells $a_t$ anchors per month, in thousands.

2. Diagram
   Observation

   An anchor drawing is labeled = $a_t$.

3. Diagram
   Observation

   The upper graph shows $a_t$ versus t with an upward trend.

4. Formula
   Observation

   The board displays $z_t = a_{t+1} - a_t$ and a lower graph labeled $z_t$.

1. No numerical data values are provided.

2. The clip does not complete a forecast calculation.

A salesman wants to predict next month's anchor sales from past monthly sales data.

1. Monthly anchor sales are denoted $a_t$.

2. t is the month or time period.

3. Sales are measured in thousands.

4. The plotted $a_t$ series has an upward linear-looking trend.

Explain why a direct ARMA approach is unsuitable and how the Integrated idea addresses the trend.

1. Expression

   $$a_t$$
   Explanation

   Plot historical monthly sales against time.

   Justification

   The speaker describes making a chart with time on the x-axis and anchors sold on the y-axis.

   Shown in the video
2. Expression

   $$\text{not stationary because the mean shifts upward}$$
   Explanation

   Recognize that the series has a moving mean.

   Justification

   The speaker explicitly says the mean is shifting upward in a linear fashion.

   Shown in the video
3. Expression

   $$\text{AR/MA/ARMA require stationarity}$$
   Explanation

   Conclude that a straight ARMA model cannot be used directly.

   Justification

   The speaker states that AR, MA, and ARMA models require stationarity.

   Shown in the video
4. Expression

   $$z_t = a_{t+1} - a_t$$
   Explanation

   Introduce a differenced series to remove the trend.

   Justification

   This formula is written on the board between the two graphs.

   Shown in the video
5. Expression

   $$\text{use ARIMA instead of plain ARMA}$$
   Explanation

   Use the Integrated component to handle the moving mean before applying ARMA-style modeling.

   Justification

   The speaker says ARIMA is used in situations where the series seems stationary except for a moving mean.

   Shown in the video

The example motivates ARIMA as ARMA plus an integration/differencing step for nonstationary series with a moving mean.

The verification is visual: the lower $z_t$ plot appears flatter than the upper $a_t$ plot, supporting the stated purpose of differencing, though no formal test is shown.

1. Audio
   Observation

   The speaker says, "So basically it's taking your number of anchors sold in thousands at some month, let's say February, minus the number of anchors sold in thousands from the previous month, say January."

2. Diagram
   Observation

   The anchor drawing labeled $q_t$ ties the symbolic series to the verbal example.

1. No numerical values are provided, only the conceptual month-to-month subtraction.

Explain what one value of the new differenced series represents in the anchor-sales example.

1. The original series counts anchors sold in thousands at each month.

2. The transformation is $z_t = q_{t+1} - q_t$.

Identify how a single $z_t$ value is produced from adjacent months.

1. Expression

   $$z_t = q_{t+1} - q_t$$
   Explanation

   Use the board formula for the transformed series.

   Justification

   The formula is written directly between the two graphs.

   Shown in the video
2. Expression

   $$$$
   Explanation

   Substitute the verbal example: February value minus January value gives one point of the new series.

   Justification

   The speaker explicitly describes February anchors minus January anchors as producing one time point of $z_t$.

   Shown in the video

One value of $z_t$ is the change in anchor sales from one month to the next, e.g. February minus January.

The answer matches both the written formula and the spoken month-to-month example.

1. Audio
   Observation

   The speaker refers to anchor sales and says we are no longer trying to predict your anchors but rather the difference between your anchor sales from one time point to the last.

2. Diagram
   Observation

   An anchor drawing labeled $q_t$ appears beside the ARIMA notation.

3. Formula
   Observation

   $Z_t = q_{t+1} - q_t$ is written on the board.

1. The numerical values of the anchor-sales series are never given.

Use an anchor-sales series $q_t$ to illustrate how ARIMA(1,1,1) is set up after differencing.

1. Original series denoted $q_t$

2. first difference defined as $Z_t = q_{t+1} - q_t$

3. model order chosen as ARIMA(1,1,1)

Express the forecasting target and write the corresponding ARIMA(1,1,1) equation for the differenced series.

1. Expression

   $$Z_t = q_{t+1} - q_t$$
   Explanation

   Replace the original level series by its first difference.

   Justification

   This is the $d=1$ step explained by the speaker.

   Shown in the video
2. Expression

   $$Z_t = \phi_1 Z_{t-1} + \theta_1 \epsilon_{t-1} + \epsilon_t$$
   Explanation

   Write the ARIMA(1,1,1) model for the differenced series using one AR lag, one MA lag, and the current error.

   Justification

   This is the equation completed on the board and verbally unpacked term by term.

   Shown in the video

The example models the differenced anchor-sales series with $Z_t = \phi_1 Z_{t-1} + \theta_1 \epsilon_{t-1} + \epsilon_t$.

The speaker verifies the setup by pointing out which term is AR, which is MA, and how the integrated part is represented by modeling $Z_t$ rather than $q_t$.

1. Audio
   Observation

   The speaker uses the example: 'we want to predict how many anchors are expected to sell next month.'

2. Diagram
   Observation

   An anchor icon is drawn on the top right, labeled '= $a_t$'.

Given historical sales data of anchors ($a_0$, ..., $a_l$) and an ARIMA(1,1,1) model fitted to the differenced sales data ($z_t$), predict the number of anchors to be sold at a future time k.

1. Historical series $a_0$, $a_1$, ..., $a_l$.

2. ARIMA(1,1,1) model for $z_t = a_{t+1} - a_t$.

3. Target time index $k > l$.

Calculate the predicted value $a_k$.

1. Expression

   $$Fit ARIMA(1,1,1) to z_t$$
   Explanation

   Use the historical data to estimate parameters $\phi_1$, $\theta_1$ and generate forecasts for future z values.

   Justification

   Standard ARIMA modeling procedure implied by the context.

   Supplementary explanation
2. Expression

   $$Forecast z_{l+1}, ..., z_{k-1}$$
   Explanation

   Generate predictions for the differenced series up to time k-1.

   Justification

   Required inputs for the recovery formula.

   Derived from the video
3. Expression

   $$Apply recovery formula: a_k = \sum_{i=1}^{k-l} z_{k-i} + a_l$$
   Explanation

   Sum the forecasted differences and add the last actual observation $a_l$.

   Justification

   Derived method for inverting differencing.

   Shown in the video

The predicted number of anchors sold at time k is $a_k$.

Check that if $k=l+1$, the formula reduces to $a_{l+1} = z_l + a_l$, which matches the definition of the difference.

1. Audio
   Observation

   Speaker identifies $q_k$ as the number of anchors sold in time period k.

2. Diagram
   Observation

   A blue anchor drawing is labeled =$q_t$.

3. Formula
   Observation

   The board uses $q_t$, $Z_t$, and ARIMA(1,1,1) throughout the anchor-sales example.

1. No numerical data values are given, so no numeric forecast is computed.

Model a time series of anchor sales $q_t$ that has a visible trend by differencing it, fitting an ARIMA(1,1,1)-style equation to the differenced series, and recovering forecasts for original sales.

1. $q_t$ denotes anchors sold at time t.

2. The plotted $q_t$ series has an upward linear trend with fluctuations.

3. $Z_t=q_{t+1}-q_t$ is the first-difference series.

4. The displayed model is ARIMA(1,1,1): $Z_t=\phi_1 Z_{t-1}+\theta_1\epsilon_{t-1}+\epsilon_t$.

Explain how ARIMA handles the trending anchor-sales series and how to recover $q_k$ from modeled differences.

1. Expression

   $$q_t \mapsto Z_t=q_{t+1}-q_t$$
   Explanation

   Convert the trending original series into a differenced series.

   Justification

   The video states first differences are used when the series has an obvious linear trend.

   Shown in the video
2. Expression

   $$Z_t=\phi_1 Z_{t-1}+\theta_1\epsilon_{t-1}+\epsilon_t$$
   Explanation

   Apply an ARMA-type equation to the differenced series.

   Justification

   This is the displayed ARIMA(1,1,1) equation.

   Shown in the video
3. Expression

   $$q_k=\sum_{i=1}^{k-1}Z_{k-i}+q_1$$
   Explanation

   Recover the original sales level by cumulatively adding predicted differences to $q_1$.

   Justification

   This is the reverse transformation written on the board.

   Shown in the video

The anchor-sales example yields a conceptual forecast procedure: difference the trending sales series, model the differences with ARIMA(1,1,1), then sum predicted differences back onto the initial sales value.

The recovery formula follows algebraically from $Z_t=q_{t+1}-q_t$; no numerical verification is present in the clip.

1. Diagram
   Observation

   Two stacked coordinate graphs are drawn on the left half of the board.

2. Formula
   Observation

   The upper vertical axis is labeled $a_t$, the lower vertical axis is labeled $z_t$, and both horizontal axes are labeled t.

3. Formula
   Observation

   A downward arrow between them is labeled $z_t = a_{t+1} - a_t$.

1. Upper graph with axis labels $a_t$ and t

2. Orange upward-trending jagged line

3. Downward arrow labeled $z_t = a_{t+1} - a_t$

4. Lower graph with axis labels $z_t$ and t

5. Purple flatter jagged line

1. The presentation moves conceptually from the upper series $a_t$ to the lower series $z_t$ via the displayed subtraction formula.

1. Both graphs share the same horizontal time axis label t.

2. The board layout remains fixed throughout the clip.

The visual contrast between the upper trending plot and the lower flatter plot illustrates the effect of first differencing.

1. Diagram
   Observation

   A blue anchor drawing appears at the top right with = $a_t$ beside it.

2. Audio
   Observation

   The speaker says the number of anchors sold every month is given by a variable a sub t.

1. Blue anchor drawing

2. Label = $a_t$

1. The anchor image is used to identify the quantity being modeled.

1. The anchor symbol consistently represents the monthly sales variable $a_t$.

The drawing provides a concrete mnemonic for the abstract time-series variable.

1. Diagram
   Observation

   The left side lists A Auto, R Regressive, I Integrated, M Moving, A Average.

2. Audio
   Observation

   The speaker explains AR and MA and later says I stands for integrated.

1. Vertical letter list A R I M A

2. Corresponding words Auto, Regressive, Integrated, Moving, Average

1. The spoken explanation aligns the letters with their expanded terms.

1. The acronym order remains A-R-I-M-A throughout.

The list visually decomposes ARIMA into its constituent model components.

1. Diagram
   Observation

   The whiteboard is organized into a left acronym column, a central pair of graphs, and a right-side ARIMA formula area.

2. Diagram
   Observation

   The upper graph is labeled $q_t$ on the vertical axis and t on the horizontal axis; the lower graph is labeled $z_t$ and t.

3. Formula
   Observation

   A downward arrow between the graphs is annotated with $z_t = q_{t+1} - q_t$.

1. Upper orange rising jagged plot labeled $q_t$ versus t

2. Lower pink fluctuating plot labeled $z_t$ versus t

3. Downward transformation arrow with $z_t = q_{t+1} - q_t$

4. Right-side anchor drawing and partial ARIMA notation

1. The presenter repeatedly points first to the upper trending plot and then to the lower differenced plot.

2. Near the end, a horizontal dashed line is added across the lower plot to indicate the level around which $z_t$ fluctuates.

1. The upper plot consistently represents the original series with an upward trend.

2. The lower plot consistently represents the transformed series after differencing.

3. The formula $z_t = q_{t+1} - q_t$ remains the explicit link between the two plots.

The visual layout teaches that ARIMA's "Integrated" part means replacing the raw series by its successive differences, and that the lower plot is meant to look more stable than the upper one.

1. Animation
   Observation

   At about 01:52 the presenter says he will use a different color and draws a horizontal dashed line across the lower graph.

2. Audio
   Observation

   He says the values "nicely hover around some mean."

1. The exact numerical level of the dashed line is not labeled.

1. Lower $z_t$ plot

2. Horizontal dashed line

1. A new dashed horizontal line is drawn through the lower plot after the verbal discussion of hovering around a mean.

1. The lower plot still represents $z_t = q_{t+1} - q_t$.

2. The line is horizontal, indicating a constant level rather than a trend.

The dashed line visually marks the constant mean that the speaker claims the differenced series now has.

1. Diagram
   Observation

   The board is divided into a left acronym column, a center pair of time-series plots, and a right ARIMA notation block with an anchor drawing.

2. Formula
   Observation

   Visible text includes ARIMA MODEL, Auto Regressive Integrated Moving Average, ARIMA(1,1,1), p,d,q, and the equation being filled in.

1. Title ARIMA MODEL

2. acronym list Auto / Regressive / Integrated / Moving / Average

3. upper plot labeled $q_t$ versus t

4. lower plot labeled $Z_t$ versus t

5. anchor drawing labeled $q_t$

6. right-side notation ARIMA(1,1,1) with p,d,q

7. equation area for $Z_t$ = ...