Exponential Smoothing in Forecasting

1. Diagram
   Observation

   The page title at the top reads "Exponential Smoothing".

Exponential Smoothing

Name of the forecasting method being introduced.

Topic label on the page

1. Audio
   Observation

   The speaker says, "the forecast in time t" while writing the left-hand side of the formula.

2. Formula
   Observation

   The written formula begins with $F_t =$.

$F_t$

Forecast at time $t$.

Time index $t$; forecast value for period $t$

1. Audio
   Observation

   The speaker says, "is equal to the forecast in time t minus one, the last forecast."

2. Formula
   Observation

   The first term on the right-hand side is written as $F_{t-1}$.

3. Diagram
   Observation

   An arrow from $F_{t-1}$ is labeled "last forecast" and "level".

$F_{t-1}$

Forecast at the previous time period $t-1$; also called the last forecast or level.

Time index $t-1$; prior-period forecast value

1. Audio
   Observation

   The speaker says, "plus alpha, which is what's called a smoothing parameter."

2. Formula
   Observation

   The coefficient multiplying the error term is written as $\alpha$.

3. Diagram
   Observation

   An arrow from $\alpha$ is labeled "smoothing parameter".

$\alpha$

Smoothing parameter that scales how much of the previous forecast error is used to adjust the next forecast.

Scalar parameter in the exponential smoothing formula

1. Audio
   Observation

   The speaker says, "actual t minus one" while writing the first term inside the parentheses.

2. Formula
   Observation

   The expression inside the parentheses begins with $A_{t-1}$.

3. Diagram
   Observation

   An arrow from $A_{t-1}$ is labeled "actual in previous period".

$A_{t-1}$

Actual observed value in the previous period $t-1$.

Time index $t-1$; realized observation or sales value

1. Audio
   Observation

   The speaker explains subtracting the forecast from the actual value and asks, "how much was I wrong by last time".

2. Formula
   Observation

   The parenthetical term is written as $(A_{t-1} - F_{t-1})$.

($A_{t-1} - F_{t-1}$)

Previous-period forecast error: actual value minus forecast value at time $t-1$.

Difference between realized value and prior forecast

1. Formula
   Observation

   Written as $F_t$ in the formula and example.

$F_t$

Forecast for period t

Real numbers

1. Formula
   Observation

   Written as $F_{t-1}$ in the formula.

$F_{t-1}$

Last forecast level (Forecast for period t-1)

Real numbers

1. Formula
   Observation

   Written as alpha in the formula and example.

$\alpha$

Smoothing parameter

Real numbers between 0 and 1

1. Formula
   Observation

   Written as $A_{t-1}$ in the formula.

$A_{t-1}$

Actual value in previous period

Real numbers

1. Formula
   Observation

   $F_t = F_{t-1} + \alpha(A_{t-1} - F_{t-1})$

$F_t$

Forecast for period t

Real numbers

1. Formula
   Observation

   $F_{t-1}$

$F_{t-1}$

Forecast for the previous period (t-1)

Real numbers

1. Audio
   Observation

   The speaker introduces the topic as "exponential smoothing" and describes it as "a more advanced sort of weighted moving average approach to forecasting".

2. Diagram
   Observation

   The page title reads "Exponential Smoothing".

In this clip, exponential smoothing is presented as an advanced weighted-moving-average style method for forecasting. The speaker contrasts it with earlier methods: naive forecasting looks one period back, simple moving average uses several past periods equally, and weighted moving average gives more weight to recent observations. Exponential smoothing is described as doing a similar kind of weighting in a different way.

1. Used for forecasting

2. Presented as related to weighted moving averages

1. Audio
   Observation

   The speaker states the notation verbally: forecast in time $t$ equals forecast in time $t-1$ plus alpha times actual $t-1$ minus forecast $t-1$.

2. Formula
   Observation

   The completed handwritten equation is $F_t = F_{t-1} + \alpha (A_{t-1} - F_{t-1})$.

3. Diagram
   Observation

   Arrows label $F_{t-1}$ as "last forecast" and "level", $\alpha$ as "smoothing parameter", and $A_{t-1}$ as "actual in previous period".

The clip writes the basic exponential smoothing recursion as a previous forecast plus a fraction of the previous forecast error. The term $F_{t-1}$ is the starting point, and $\alpha(A_{t-1}-F_{t-1})$ adjusts that old forecast using the error from the last period.

1. $F_t$ is the forecast for time $t$

2. $F_{t-1}$ is the forecast made for the previous period

3. $A_{t-1}$ is the actual observed value in the previous period

4. $\alpha$ is the smoothing parameter

1. Exponential smoothing as a forecasting method

1. Audio
   Observation

   The speaker calls $\alpha$ "a smoothing parameter" and later says, "So if alpha is high, you give more..." before the clip ends.

2. Formula
   Observation

   $\alpha$ multiplies the error term $(A_{t-1} - F_{t-1})$ in the written equation.

3. Diagram
   Observation

   An arrow labels $\alpha$ as "smoothing parameter".

1. The sentence explaining the effect of a high $\alpha$ is cut off before completion.

The smoothing parameter $\alpha$ controls how strongly the previous forecast error affects the updated forecast. In the written formula, it multiplies the error term, so larger $\alpha$ means a larger correction carried forward from the last period’s mistake.

1. $\alpha$ is applied to the previous forecast error

2. The clip does not state the full numerical range of $\alpha$

1. One-step exponential smoothing update formula

1. Audio
   Observation

   The speaker says the method takes actual sales from the last period, subtracts the forecast, asks "how much was I wrong by last time," and then adjusts the forecast partially to reflect that error.

2. Formula
   Observation

   The subtraction appears explicitly as $(A_{t-1} - F_{t-1})$.

The clip interprets the update rule operationally: compute the previous period’s error, then move the next forecast partway toward correcting that error. This makes the formula understandable as a partial adjustment rather than a full replacement of the old forecast.

1. Applies to one-period-ahead updating

2. Uses the previous actual value and previous forecast

1. One-step exponential smoothing update formula
2. Role of the smoothing parameter alpha

1. Formula
   Observation

   $F_t = F_{t-1} + \alpha(A_{t-1} - F_{t-1})$

2. Audio
   Observation

   The speaker explains the components of the formula.

The formula for exponential smoothing is $F_t = F_{t-1} + \alpha(A_{t-1} - F_{t-1})$. It updates the forecast for the current period ($F_t$) by taking the previous forecast ($F_{t-1}$) and adding a fraction ($\alpha$) of the error from the previous period ($A_{t-1} - F_{t-1}$).

1. $\alpha$ is the smoothing parameter.

2. $F_{t-1}$ is the last forecast level.

3. $A_{t-1}$ is the actual in the previous period.

1. Formula
   Observation

   alpha is high -> give more weight to recent data.

2. Audio
   Observation

   The speaker explains that a higher alpha gives more weight to recent sales/data.

When the smoothing parameter $\alpha$ is high, the model gives more weight to recent data. The forecast responds more significantly to the difference between the actual and forecasted values in the previous period.

1. $\alpha$ is close to 1.

1. Exponential Smoothing Formula

1. Formula
   Observation

   alpha is low -> give more weight to historical data.

2. Audio
   Observation

   The speaker explains that a lower alpha gives more weight to historical data.

When the smoothing parameter $\alpha$ is low, the model gives more weight to historical data. The forecast responds less to the recent error, relying more on the average of many previous sales.

1. $\alpha$ is close to 0.

1. Exponential Smoothing Formula

1. Formula
   Observation

   $F_t = F_{t-1} + \alpha(A_{t-1} - F_{t-1})$

2. Audio
   Observation

   this is just the formula plus alpha times A t minus one minus F t minus one

The formula used to calculate the forecast for a new period based on the previous period's forecast and actual value.

1. Requires an initial forecast value.

2. Requires a smoothing constant alpha.

1. Formula
   Observation

   Handwritten formula on the paper: $F_t = F_{t-1} + α(A_{t-1} - F_{t-1})$.

2. Audio
   Observation

   The speaker explains that you start with the trend and adjust using the smoothing parameter and the difference between the actual and the forecast.

The clip presents exponential smoothing as a recursive forecast update: take the previous forecast and add a fraction α of the previous forecast error. The error term is the actual value minus the previous forecast.

1. The formula is applied recursively from one period to the next.

2. The worked example uses the same initial forecast value 20 for week 2 across all α columns.

3. The video does not state a formal domain restriction on α.

1. Audio
   Observation

   The speaker says, "as alpha gets bigger, you respond to the more recent one more significantly."

2. Audio
   Observation

   The speaker also says, "the bigger alpha is, the more weight you put on the most recent sales."

3. Diagram
   Observation

   Three forecast columns are labeled $α=0.1$, $α=0.3$, and $α=0.5$, showing how changing α changes the forecast sequence.

In this lesson, α is the smoothing parameter that determines how strongly the forecast responds to the latest actual observation. Larger α gives more weight to the most recent actual value and makes the forecast more responsive.

1. The explanation is based on the worked comparison among $α=0.1$, $α=0.3$, and $α=0.5$.

2. The clip does not define α by an explicit weighted-average expansion beyond the recurrence.

1. Exponential smoothing recurrence

1. Audio
   Observation

   The speaker states, "So essentially exponential smoothing becomes a weighted moving average."

2. Audio
   Observation

   He continues, "and the bigger alpha is, the more weight you put on the most recent sales."

The instructor interprets exponential smoothing as a weighted moving average in which the newest actual observation receives larger weight when α is larger.

1. This is presented as an interpretation of the recurrence rather than a separate derivation.

2. The clip does not expand the infinite weighted form explicitly.

1. Exponential smoothing recurrence
2. Role of the smoothing parameter α

1. Audio
   Observation

   The speaker says, "you use one and two for ... or you use two for three, use three for four, and so on and so on."

2. Diagram
   Observation

   Arrows drawn near the end connect row 2 to row 3 and row 3 to row 4, visually indicating the lagged use of prior-period values.

1. The spoken phrase around "you use one and two for" is slightly disfluent, but the intended indexing rule is clear from the rest of the sentence and the arrows.

To compute the forecast for a given week, the method uses the actual value and forecast from the immediately preceding week. The instructor emphasizes keeping straight which period is being forecast and which prior numbers feed the update.

1. Applies to the recursive table calculation shown for weeks 2 through 6.

2. The video demonstrates the rule with arrows linking adjacent rows.

1. Exponential smoothing recurrence

1. Audio
   Observation

   The speaker says, "as alpha gets bigger, you respond to the more recent one more significantly."

2. Audio
   Observation

   He adds, "the bigger alpha is, the more weight you put on the most recent sales."

3. Diagram
   Observation

   The three columns compare forecasts for $α=0.1$, $α=0.3$, and $α=0.5$ using the same actual data.

For the same sequence of actual observations, increasing α places more weight on the most recent actual value and makes the forecast respond more strongly to it.

1. The comparison is made within the exponential-smoothing recurrence shown in the clip.

2. The same initial forecast and actual series are used across the compared α columns.

For the worked example and the instructor's general verbal claim, larger α corresponds to stronger response to the most recent observation.

1. Audio
   Observation

   The speaker says, "So essentially exponential smoothing becomes a weighted moving average."

1. The clip states this as an interpretive claim without deriving the weighted-sum representation.

Exponential smoothing can be understood as a weighted moving average of past observations, with weights determined by α.

1. The claim is made in the context of the recurrence $F_t = F_{t-1} + α(A_{t-1} - F_{t-1})$.

Stated generally by the instructor for exponential smoothing, though not formally proved in this clip.

1. Audio
   Observation

   The speaker builds the formula aloud piece by piece: first $F_t$, then $F_{t-1}$, then $+\alpha$, then the parenthetical error term.

2. Formula
   Observation

   The handwriting progresses from $F_t = F_{t-1}$ to $F_t = F_{t-1} + \alpha$ and finally to $F_t = F_{t-1} + \alpha (A_{t-1} - F_{t-1})$.

3. Diagram
   Observation

   Labels are added under or beside terms as they are introduced.

1. Expression

   $$F_t$$
   Explanation

   Start with the quantity being defined: the forecast at time $t$.

   Justification

   The speaker says, "the forecast in time t" while writing the left-hand side.

   Shown in the video
2. Expression

   $$F_t = F_{t-1}$$
   Explanation

   Begin the update from the previous forecast.

   Justification

   The speaker says this equals "the forecast in time t minus one, the last forecast."

   Shown in the video
3. Expression

   $$F_t = F_{t-1} + \alpha$$
   Explanation

   Add a correction term scaled by the smoothing parameter.

   Justification

   The speaker says "plus alpha, which is what's called a smoothing parameter."

   Shown in the video
4. Expression

   $$F_t = F_{t-1} + \alpha (A_{t-1} - F_{t-1})$$
   Explanation

   Multiply $\alpha$ by the previous period’s forecast error, actual minus forecast.

   Justification

   The speaker writes and explains "actual t minus one minus forecast t minus one" and interprets it as "how much was I wrong by last time."

   Shown in the video

The displayed recursion updates the old forecast by adding a fraction $\alpha$ of the previous forecast error.

1. Formula
   Observation

   $F_2 = 20 + \alpha(20 - 20) = 20$

2. Audio
   Observation

   forecast for period two is equal to the forecast for period one which is 20 plus alpha times 20 minus 20 so it's going to equal to 20

1. Expression

   $$F_2 = 20 + \alpha(20 - 20)$$
   Explanation

   Substitute the given values into the exponential smoothing formula.

   Justification

   Given $F_1 = 20$ and $A_1 = 20$.

   Shown in the video
2. Expression

   $$F_2 = 20$$
   Explanation

   Simplify the expression.

   Justification

   Arithmetic calculation.

   Shown in the video

$F_2 = 20$

1. Formula
   Observation

   $F_3 = 20 + 0.1(22 - 20) = 20 + 0.2 = 20.2$

2. Audio
   Observation

   Forecast for period three is equal to 20, which is the forecast for period two, plus 0.1, which is the alpha that we were given, 22 was the actual in period two, minus 20, which was our forecast, equals 20 plus 0.2, that difference times 0.1, equals 20.2.

1. Expression

   $$F_3 = 20 + 0.1(22 - 20)$$
   Explanation

   Substitute the calculated $F_2$, given alpha, and given $A_2$ into the formula.

   Justification

   $F_2 = 20$, $\alpha = 0.1$, $A_2 = 22$.

   Shown in the video
2. Expression

   $$F_3 = 20 + 0.2$$
   Explanation

   Calculate the difference and multiply by alpha.

   Justification

   Arithmetic calculation.

   Shown in the video
3. Expression

   $$F_3 = 20.2$$
   Explanation

   Add the terms.

   Justification

   Arithmetic calculation.

   Shown in the video

$F_3 = 20.2$

1. Formula
   Observation

   $F_4 = 20.2 + 0.1(18 - 20.2) = 19.98$

2. Audio
   Observation

   Forecast for period four is equal to 20.2 plus 0.1 times 18 minus 20.2. Actual is lower, so this is going to be a negative number multiplied by 0.1, sorry, and then this number is going to get lower, so we're going to get 19.98.

1. Expression

   $$F_4 = 20.2 + 0.1(18 - 20.2)$$
   Explanation

   Substitute the calculated $F_3$, given alpha, and given $A_3$ into the formula.

   Justification

   $F_3 = 20.2$, $\alpha = 0.1$, $A_3 = 18$.

   Shown in the video
2. Expression

   $$F_4 = 19.98$$
   Explanation

   Calculate the result.

   Justification

   Arithmetic calculation.

   Shown in the video

$F_4 = 19.98$

1. Formula
   Observation

   $F_5 = 19.98 + 0.1(21 - 19.98) = 20.1$

2. Audio
   Observation

   The forecast for period five then is equal to 19.98 plus 0.1 times 21 minus 19.98 equals 20.1

1. Expression

   $$F_5 = 19.98 + 0.1(21 - 19.98)$$
   Explanation

   Substitute the calculated $F_4$, given alpha, and given $A_4$ into the formula.

   Justification

   $F_4 = 19.98$, $\alpha = 0.1$, $A_4 = 21$.

   Shown in the video
2. Expression

   $$F_5 = 20.1$$
   Explanation

   Calculate the result.

   Justification

   Arithmetic calculation.

   Shown in the video

$F_5 = 20.1$

1. Formula
   Observation

   $F_6 = 20.3$

2. Audio
   Observation

   and the forecast for period six would be easy to calculate is 20.3.

1. The intermediate calculation steps for $F_6$ are not shown or spoken.

1. Expression

   $$F_6 = 20.3$$
   Explanation

   State the final calculated value.

   Justification

   Stated directly by the speaker without showing the substitution and arithmetic steps.

   Shown in the video

$F_6 = 20.3$

1. Formula
   Observation

   The paper sets $F2=20$ and labels the first forecast column $α=0.1$.

2. Diagram
   Observation

   The $α=0.1$ column contains 20, 20.2, 19.98, 20.1, 20.3 for weeks 2 through 6.

3. Audio
   Observation

   The speaker says, "So this is for alpha equals one."

1. The clip does not show every intermediate arithmetic line for the $α=0.1$ column; only the final column values are visible.

1. Expression

   $$F_2 = 20$$
   Explanation

   The forecast for week 2 is initialized at 20 in the $α=0.1$ column.

   Justification

   Visible table entry under $α=0.1$.

   Shown in the video
2. Expression

   $$F_3=20+0.1(22-20)=20.2$$
   Explanation

   Using the week-2 actual 22 and the week-2 forecast 20, the next forecast becomes 20.2.

   Justification

   Matches the recurrence and the visible $α=0.1$ column value for week 3.

   Derived from the video
3. Expression

   $$F_4=20.2+0.1(18-20.2)=19.98$$
   Explanation

   Using the week-3 actual 18 and the week-3 forecast 20.2, the next forecast becomes 19.98.

   Justification

   Matches the recurrence and the visible $α=0.1$ column value for week 4.

   Derived from the video
4. Expression

   $$F_5=19.98+0.1(21-19.98)=20.082\approx20.1$$
   Explanation

   Using the week-4 actual 21 and the week-4 forecast 19.98, the next forecast becomes 20.1.

   Justification

   Matches the recurrence and the visible $α=0.1$ column value for week 5.

   Derived from the video
5. Expression

   $$F_6=20.082+0.1(22-20.082)=20.2738\approx20.3$$
   Explanation

   Using the week-5 actual 22 and the week-5 forecast 20.1, the next forecast becomes 20.3.

   Justification

   Matches the recurrence and the visible $α=0.1$ column value for week 6.

   Derived from the video

With $α=0.1$, the table gives 20,20.2,19.98,20.1,20.3 after display rounding; $α=1$ would instead copy the previous actual exactly.

1. Diagram
   Observation

   The $α=0.3$ column is filled with 20.6, 19.8, 20.2, 20.7 for weeks 3 through 6.

2. Audio
   Observation

   The speaker says, "For alpha is equal to 0.3, all we would change is that this would go from 0.1 to 0.3," then reads the resulting values.

1. The clip does not display the full intermediate algebra for each row in the $α=0.3$ column.

1. Expression

   $$F_2 = 20$$
   Explanation

   The starting forecast for week 2 is again 20.

   Justification

   Visible table entry under $α=0.3$.

   Shown in the video
2. Expression

   $$F_3 = 20 + 0.3(22-20) = 20.6$$
   Explanation

   The week-3 forecast is obtained by adding 30% of the week-2 forecast error to the week-2 forecast.

   Justification

   Matches the recurrence and the visible $α=0.3$ column value for week 3.

   Derived from the video
3. Expression

   $$F_4 = 20.6 + 0.3(18-20.6) = 19.8$$
   Explanation

   The week-4 forecast uses the week-3 actual 18 and the week-3 forecast 20.6.

   Justification

   Matches the recurrence and the visible $α=0.3$ column value for week 4.

   Derived from the video
4. Expression

   $$F_5 = 19.8 + 0.3(21-19.8) = 20.2$$
   Explanation

   The week-5 forecast uses the week-4 actual 21 and the week-4 forecast 19.8.

   Justification

   Matches the recurrence and the visible $α=0.3$ column value for week 5.

   Derived from the video
5. Expression

   $$F_6 = 20.2 + 0.3(22-20.2) = 20.7$$
   Explanation

   The week-6 forecast uses the week-5 actual 22 and the week-5 forecast 20.2.

   Justification

   Matches the recurrence and the visible $α=0.3$ column value for week 6.

   Derived from the video

With $α=0.3$, the forecast adjusts by 30% of the previous forecast error each period, giving the column 20, 20.6, 19.8, 20.2, 20.7.

1. Diagram
   Observation

   The source writes 21,19.3,20.3,21.1 in the $α=0.5$ column; the week-4 value is incorrect.

2. Audio
   Observation

   The narration reads the same source values; independent substitution gives week $4 = 19.5$.

1. The clip does not display the full intermediate algebra for each row in the $α=0.5$ column.

1. Expression

   $$F_2 = 20$$
   Explanation

   The starting forecast for week 2 is 20.

   Justification

   Visible table entry under $α=0.5$.

   Shown in the video
2. Expression

   $$F_3 = 20 + 0.5(22-20) = 21$$
   Explanation

   The week-3 forecast adds half of the week-2 forecast error to the week-2 forecast.

   Justification

   Matches the recurrence and the visible $α=0.5$ column value for week 3.

   Derived from the video
3. Expression

   $$F_4=21+0.5(18-21)=19.5$$
   Explanation

   The week-4 forecast uses the week-3 actual 18 and the week-3 forecast 21.

   Justification

   Matches the recurrence and the visible $α=0.5$ column value for week 4.

   Derived from the video
4. Expression

   $$F_5=19.5+0.5(21-19.5)=20.25\approx20.3$$
   Explanation

   Using the corrected week-4 forecast gives 20.25, displayed to one decimal as 20.3.

   Justification

   Matches the recurrence and the visible $α=0.5$ column value for week 5.

   Derived from the video
5. Expression

   $$F_6=20.25+0.5(22-20.25)=21.125\approx21.1$$
   Explanation

   The week-6 forecast uses the week-5 actual 22 and the week-5 forecast 20.3.

   Justification

   Matches the recurrence and the visible $α=0.5$ column value for week 6.

   Derived from the video

Corrected $α=0.5$ forecasts are 20,21,19.5,20.25,21.125, or 20,21,19.5,20.3,21.1 at the video’s display precision. The source writes 19.3 for week 4 by mistake.

1. Diagram
   Observation

   Near the end of the clip, arrows are drawn from row 2 toward row 3 and from row 3 toward row 4.

2. Audio
   Observation

   The speaker says, "you use two for three, use three for four, and so on and so on."

1. Expression

   $$F_3 \leftarrow (A_2, F_2)$$
   Explanation

   The arrow from week 2 to week 3 indicates that the forecast for week 3 is computed from the actual and forecast of week 2.

   Justification

   Visible arrow plus spoken indexing rule.

   Shown in the video
2. Expression

   $$F_4 \leftarrow (A_3, F_3)$$
   Explanation

   The arrow from week 3 to week 4 indicates that the forecast for week 4 is computed from the actual and forecast of week 3.

   Justification

   Visible arrow plus spoken indexing rule.

   Shown in the video
3. Expression

   $$F_t \leftarrow (A_{t-1}, F_{t-1})$$
   Explanation

   The pattern generalizes to each later period using the immediately preceding row.

   Justification

   Spoken phrase "and so on and so on" following the demonstrated arrows.

   Derived from the video

The arrows encode the recursive dependence of each forecast on the previous period's actual value and previous period's forecast.

1. Formula
   Observation

   Table with Weeks 1-6 and Actual values 20, 22, 18, 21, 22.

2. Formula
   Observation

   Exponential smoothing forecast for $1 = 20$

3. Formula
   Observation

   forecast using alpha = 0.1

4. Audio
   Observation

   The speaker reads out the data and sets up the problem.

1. The calculation is cut off before completion.

Calculate the exponential smoothing forecast using the provided weekly actual data, a starting forecast of 20 for week 1, and a smoothing parameter $\alpha = 0.1$.

1. Weeks: 1, 2, 3, 4, 5, 6

2. Actuals: 20, 22, 18, 21, 22

3. Starting forecast for week 1 ($F_1$) = 20

4. Smoothing parameter ($\alpha$) = 0.1

Compute the forecasts for subsequent weeks.

1. Expression

   $$F_1 = 20$$
   Explanation

   Set the initial forecast for week 1.

   Justification

   Given in the problem setup.

   Shown in the video
2. Expression

   $$\alpha = 0.1$$
   Explanation

   Set the smoothing parameter.

   Justification

   Given in the problem setup.

   Shown in the video
3. Expression

   $$F_t = F_{t-1} + \alpha(A_{t-1} - F_{t-1})$$
   Explanation

   Write down the formula to be used for calculation.

   Justification

   Definition of exponential smoothing.

   Shown in the video

Incomplete calculation shown in the video.

Not shown in the video.

1. Diagram
   Observation

   Table with 'Weeks' and 'Actual' columns, and a third column being filled with forecasts.

2. Audio
   Observation

   Exponential smoothing forecast for... forecast using alpha = 0.1

Calculate the exponential smoothing forecasts for weeks 2 through 6, given the actual values for weeks 1 through 5, an initial forecast for week 1, and a smoothing constant alpha = 0.1.

1. Initial forecast $F_1 = 20$

2. Actual values: $A_1 = 20$, $A_2 = 22$, $A_3 = 18$, $A_4 = 21$, $A_5 = 22$

3. Smoothing constant $\alpha = 0.1$

Find $F_2$, $F_3$, $F_4$, $F_5$, and $F_6$.

1. Expression

   $$F_2 = 20 + 0.1(20 - 20) = 20$$
   Explanation

   Calculate forecast for week 2.

   Justification

   Using the exponential smoothing formula with $F_1$ and $A_1$.

   Shown in the video
2. Expression

   $$F_3 = 20 + 0.1(22 - 20) = 20.2$$
   Explanation

   Calculate forecast for week 3.

   Justification

   Using the formula with $F_2$ and $A_2$.

   Shown in the video
3. Expression

   $$F_4 = 20.2 + 0.1(18 - 20.2) = 19.98$$
   Explanation

   Calculate forecast for week 4.

   Justification

   Using the formula with $F_3$ and $A_3$.

   Shown in the video
4. Expression

   $$F_5 = 19.98 + 0.1(21 - 19.98) = 20.1$$
   Explanation

   Calculate forecast for week 5.

   Justification

   Using the formula with $F_4$ and $A_4$.

   Shown in the video
5. Expression

   $$F_6 = 20.3$$
   Explanation

   State forecast for week 6.

   Justification

   Calculated off-screen or mentally by the speaker.

   Shown in the video

$F_2 = 20$, $F_3 = 20.2$, $F_4 = 19.98$, $F_5 = 20.1$, $F_6 = 20.3$

The calculated values are written into the table on the paper.

1. Diagram
   Observation

   A handwritten table lists Weeks 1–6, Actual values 20, 22, 18, 21, 22, and three forecast columns labeled $α=0.1$, $α=0.3$, and $α=0.5$.

2. Audio
   Observation

   The speaker reads and compares the forecast values for the different α choices.

3. Formula
   Observation

   The recurrence $F_t = F_{t-1} + α(A_{t-1} - F_{t-1})$ is written below the table.

1. The clip does not state the real-world source of the sales numbers beyond them being weekly sales.

Given actual sales 20,22,18,21,22 for weeks 1–5 and an initial week-2 forecast of 20, compare forecasts for $α=0.1$, $α=0.3$ and $α=0.5$.

1. Actual sales: week $1 = 20$, week $2 = 22$, week $3 = 18$, week $4 = 21$, week $5 = 22$.

2. Initial forecast for week 2 is 20 in each α column.

3. Forecasting rule: $F_t = F_{t-1} + α(A_{t-1} - F_{t-1})$.

4. Comparison values used: $α=0.1$, $α=0.3$, $α=0.5$.

Fill the forecast table for weeks 2 through 6 and compare how different α values change the forecasts.

1. Expression

   $$\alpha=0.1:\quad 20,\ 20.2,\ 19.98,\ 20.1,\ 20.3$$
   Explanation

   The $α=0.1$ column is least responsive to recent actuals among the three shown.

   Justification

   Visible table entries under $α=0.1$.

   Shown in the video
2. Expression

   $$\alpha=0.3:\quad 20,\ 20.6,\ 19.8,\ 20.2,\ 20.7$$
   Explanation

   The $α=0.3$ column is then written using the same recurrence with a smaller smoothing weight.

   Justification

   Visible table entries under $α=0.3$ and spoken explanation that only α changes.

   Shown in the video
3. Expression

   $$\alpha=0.5:\quad 20,\ 21,\ 19.5,\ 20.3,\ 21.1$$
   Explanation

   The corrected $α=0.5$ column replaces the source’s erroneous 19.3 with 19.5; later displayed values match full-precision recursion.

   Justification

   Editorial correction of the displayed $α=0.5$ week-4 arithmetic.

   Shown in the video
4. Expression

   $$F_3 \text{ uses } A_2,F_2;\quad F_4 \text{ uses } A_3,F_3$$
   Explanation

   At the end, arrows are drawn to emphasize that each forecast uses the immediately preceding row.

   Justification

   Visible arrows and spoken indexing reminder.

   Shown in the video

Corrected comparison: $α=0.1$: 20,20.2,19.98,20.1,20.3; $α=0.3$: 20,20.6,19.8,20.2,20.7; $α=0.5$: 20,21,19.5,20.3,21.1. The source shows 19.3 instead of 19.5 at week 4.

Independent recurrence using unrounded values confirms week-4 $α=0.5$ forecast 19.5, followed by 20.25 and 21.125.

1. Diagram
   Observation

   A sheet titled "Exponential Smoothing" is shown while the speaker talks and gestures with a pen.

1. Handwritten title "Exponential Smoothing"

2. Pen

3. Blank paper below the title

1. The speaker gestures over the blank area before writing the formula.

1. The page title remains visible at the top.

This visual establishes the topic before any notation is introduced.

1. Diagram
   Observation

   The formula is written progressively and annotated with arrows and labels.

2. Formula
   Observation

   The final visible equation is $F_t = F_{t-1} + \alpha (A_{t-1} - F_{t-1})$.

1. Equation $F_t = F_{t-1} + \alpha (A_{t-1} - F_{t-1})$

2. Arrow from $F_{t-1}$ labeled "last forecast" and "level"

3. Arrow from $\alpha$ labeled "smoothing parameter"

4. Arrow from $A_{t-1}$ labeled "actual in previous period"

1. First $F_t = F_{t-1}$ is written.

2. Then $+\alpha$ is added.

3. Then the parenthetical error term $(A_{t-1} - F_{t-1})$ is completed.

4. Labels are added beneath or beside the corresponding terms.

1. The main equation remains centered on the page once written.

2. Each annotation points to the term it explains.

The visual annotations map each symbol to its forecasting meaning: prior forecast, smoothing weight, and prior actual value.

1. Animation
   Observation

   Hand writing the exponential smoothing formula and explaining the effect of alpha on a piece of paper.

1. Paper

2. Pen

3. Hand

4. Formula text

1. Text is written on the paper explaining the formula and the effects of high and low alpha.

1. The camera angle remains fixed on the paper.

Visual demonstration of the mathematical formula and its conceptual components.

1. Animation
   Observation

   Hand removes the first piece of paper and places a blank one.

1. Two pieces of paper

2. Hands

1. The first paper is removed, and a blank paper is placed on the table.

1. The camera angle remains fixed.

Transition from theoretical explanation to practical application.

1. Animation
   Observation

   Hand writing a table of weeks and actual data, and setting up the problem parameters.

1. Blank paper

2. Pen

3. Hand

4. Table text

1. A table with 'Weeks' and 'Actual' columns is drawn and filled with data. The problem parameters are written below.

1. The camera angle remains fixed on the paper.

Visual setup of a numerical example for exponential smoothing.

1. Animation
   Observation

   Hand writes calculated forecast values into the third column of the table.

1. Paper with a table

2. Pen

3. Hand

1. Empty cells in the third column are filled with numerical values (20, 20.2, 19.98, 20.1, 20.3).

1. The structure of the table remains the same.

2. The 'Weeks' and 'Actual' columns remain unchanged.

The visual action demonstrates the step-by-step application of the exponential smoothing formula to generate a sequence of forecasts.

1. Animation
   Observation

   The source fills $α=0.1$, then $α=0.3$, then $α=0.5$; its $α=0.5$ week-4 value 19.3 should be 19.5.

2. Diagram
   Observation

   The paper shows a structured table with Weeks, Actual, and three α-labeled forecast columns.

1. Handwritten table with Weeks and Actual columns

2. Forecast columns labeled $α=0.1$, $α=0.3$, $α=0.5$

3. Pen writing numerical entries

1. The $α=0.1$ column is completed first.

2. The $α=0.3$ column is added next.

3. The $α=0.5$ column is added last.

4. The final view shows all three forecast sequences side by side for comparison.

1. The Weeks and Actual columns remain fixed while the forecast columns are filled.

2. The same recurrence formula is used across all three columns.

3. The initial forecast value 20 for week 2 is common to all three columns.

The visual layout lets the viewer compare how different smoothing parameters produce different forecast trajectories from the same actual sales history.

1. Animation
   Observation

   Near the end, arrows are drawn from row 2 to row 3 and from row 3 to row 4.

2. Audio
   Observation

   The speaker explains, "you use two for three, use three for four, and so on and so on."