CHAPTER 3 ASSIGNMENT

GROUP 3
Team member:
Trần Khải Như - 2252594
Lại Khánh Vy - 2214028
Phan Đức Tuấn - 2252875
Đoàn Nam Trung - 2252852

Example 1:
Data about the quantity of VCRs sold by Vernon's Music Store in the previous year
has been gathered by Harry Vernon. The information is displayed in Table 1. The
calculations that result in the lag 1 autocorrelation coefficient's calculation are
displayed in Table 2. A scatter plot of the observation pairs (Yt, Yt-1) can be found
in Figure 4. The scatter plot makes it very evident that there will be a positive lag 1
connection. This time series has positive lag 1 autocorrelation, as the figure in
Figure 4 suggests.
For time lag 1, there is a 0.572 correlation between Yt and the autocorrelation.This
indicates that there is some correlation between the subsequent monthly sales of
VCRs.

Harry may learn a lot from this knowledge about his time series, get ready to
utilize an advanced forecasting technique, and be cautioned against doing
regression analysis on his data.
The second-order autocorrelation coefficient (r2), or the correlation between Yt
and for Harry’s data, is also computed using Equation:

This time series looks to have moderate autocorrelation that is lagging two time
periods.Yt and, or the autocorrelation for time lag 2, have a correlation of.463. It is
evident that there is a decrease in the autocorrelation coefficient at time lag 2
(.463) compared to time lag 1 (.572). In general, the magnitudes of the
autocorrelation coefficients drop as the number of time delays (k) rises.
Plotting the autocorrelations against time delays for the Harry Vernon data used in
Example 1 is presented in Figure 5. Every relevant time lag is displayed on the
graph's horizontal scale, which goes from 1 to 3.The potential range of an
autocorrelation coefficient, from -1 to 1, is displayed on the vertical scale on the
left. The graph's central horizontal line denotes autocorrelations of zero.The
autocorrelation coefficient of the vertical line rising above time lag 1 is.57, or. The
autocorrelation coefficient of the vertical line rising above time lag 2 is.46, or r2
=.46.

Example 2:
Problem:
- Determine if there's a significant autocorrelation at lag 1 in a given time series.
- Use a 5% significance level.
Steps:
- State hypotheses:
Null hypothesis (H₀): There is no autocorrelation at lag 1.
Alternative hypothesis (H₁): There is autocorrelation at lag 1.
- Calculate test statistic: Use the formula t = r₁ / SE(r₁), where r₁ is the
autocorrelation coefficient at lag 1 and SE(r₁) is its standard error.
Determine critical values: Find the critical values for a t-distribution with n-1
degrees of freedom and the desired significance level.
Make a decision: If the absolute value of the test statistic is greater than the critical
value, reject the null hypothesis and conclude there is significant autocorrelation.
Conclusion: Based on the calculated test statistic and critical values, determine
whether there is evidence of autocorrelation at lag 1 in the time series.

Example 3:
A time series of 40 random numbers was generated using Minitab. The data were
plotted, and the autocorrelation function was constructed. The autocorrelations for
the first 10 lags were examined, and all were within the 95% confidence limits,
indicating no significant autocorrelation. The Ljung-Box Q statistic was used to
test the hypothesis of no autocorrelation, and the result supported the randomness
of the data. This example demonstrates how to check for randomness in a time
series using autocorrelation and the Ljung-Box Q statistic.

Problem 7:
a. A series that has a trend: non-stationary because the trend indicates a change
over time.
b. A series whose mean and variance remain constant over time: Stationary
because both the mean and variance do not change.
c. A series whose mean value is changing over time: Non-stationary because a
changing mean indicates a shift in the series.
d. A series that contains no growth or decline: Stationary because it indicates
stability without trends.

Problem 8:
a. The series has basic statistical properties, such as the mean and variance, that
remain constant over time: Stationary.
b. The successive values of a time series are not related to each other: Random.
c. A high relationship exists between each successive value of a series: Trending.
d. A significant autocorrelation coefficient appears at time lag 4 for quarterly data:
Seasonal.
e. The series contains no growth or decline: Stationary.
f. The autocorrelation coefficients are typically significantly different from zero for
the first several time lags and then gradually decrease toward zero as the number of
lags increases: Trending.

Problem 9:
 Naive: If the stock closed at $100 this month, the naïve forecast for next one
would be $100
 Seasonalty: Sales are typically higher in the winter months and lower in the
summer months. The forecast for December would be based on the average sales
in December of previous years
 Moving average: Calculate the average daily temperature for the past 30 days to
forecast the average temperature for the next month
 Exponetial smoothing: Forecasting monthly unemployment rates. Assign a higher
weight to recent unemployment rates when calculating the forecast. This can help
capture economic trends or seasonal fluctuations
 ARIMA: GDP growth might depend on its past values (AR). If GDP growth is
non-stationary (has a trend), differencing can be used to make it stationary (I). The
current GDP growth might depend on past errors in the forecast (MA)

Problem 10:
Forecasting techniques when forecasting a trending series:
 Naïve:

 Moving averages

 Simple regression

From these data, we can find equation: Sales = 69.3333 + 21.1429 x days
For instance, we want to find sales of day 9 = 69.3333 + 21.1429 x 9 ≈ 260
 Growth curves:
 Source: Population Division of the Department of Economic and Social Affairs of
the United Nations Secretariat
 Autoregressive integrated moving average (ARIMA) models (Box-Jenkins): It is
best suited for forecasting within time frames of 18 months or less. One use for
Box-Jenkins Model analysis is to forecast stock prices

Problem 11:
Seasonal decomposition
Description:
This technique involves decomposing the time series into trend, seasonal, and
residual (noise) components. The seasonal component is isolated, making it easier
to identify and forecast recurring patterns.
Example:
Retail sales data often exhibit strong seasonal patterns, such as increased sales
during the holiday season. Seasonal decomposition can help forecast future sales
by accounting for these recurring seasonal spikes.
Exponential Smoothing (Holt-Winters method)
Description:
The Holt-Winters method extends exponential smoothing to deal with both trend
and seasonality. There are two variations: additive (for constant seasonal
variations) and multiplicative (for proportional seasonal variations).
Example:
Forecasting electricity demand often involves seasonality (higher usage in winter
or summer). Holt-Winters is ideal for capturing both the seasonal peaks and the
trend in overall demand growth.
Seasonal Autoregressive Integrated Moving Average (SARIMA)
Description:
SARIMA extends ARIMA by including seasonal components. It captures the
relationship between past values (autoregressive terms), past forecast errors
(moving average terms), and any seasonal repeating structure.
Example:
SARIMA could be used to forecast monthly airline passenger numbers, which
typically display strong seasonal patterns (e.g., higher travel in summer and during
holidays).
Fourier Transform (Fourier Analysis)
Description:
Fourier analysis transforms time series data into the frequency domain, allowing
the identification of repeating patterns over time. This is particularly useful for
identifying multiple overlapping seasonalities.
Example:
Oceanographic data such as tidal patterns and wave heights are highly seasonal and
cyclical. Fourier analysis can be applied to forecast these patterns based on the
frequency of cycles.
Time Series Regression Models with Seasonal Dummies
Description:
This method uses regression models with dummy variables to account for seasonal
effects. Each season or time period (e.g., month or quarter) is assigned a dummy
variable, which helps in capturing seasonal impacts.
Example:
A company forecasting quarterly financial earnings may use dummy variables to
account for seasonal fluctuations in revenue (e.g., higher revenue in Q4 due to
holiday shopping).

Problem 12:
- Moving Averages: This technique smooths out short-term fluctuations and
highlights longer-term trends or cycles. Ex: Forecasting monthly sales for a retail
store that experiences seasonal fluctuations and cyclical trends, such as increased
sales during holiday seasons.
- Seasonal Decomposition of Time Series (STL): Decomposes a series into
seasonal, trend, and residual components, allowing for better analysis of cyclical
patterns. Ex: Analyzing and forecasting the cyclical demand for agricultural
products influenced by seasonal planting and harvesting periods.
- Regression Analysis: Models the relationship between a dependent variable and
one or more independent variables to capture cyclical behavior. Ex: Forecasting
stock prices based on economic indicators that exhibit cyclical movements, such as
interest rates and inflation.
Problem 15:
a. Mean Error: This measure calculates the average of the forecast errors
b. Mean Absolute Percentage Error (MAPE): MAPE expresses forecast accuracy
as a percentage of the actual values, making it scale-independent and allowing for
easy comparison across different magnitudes of the forecast variable.
c. Mean Squared Error (MSE) or Root Mean Squared Error (RMSE): oth measures
square the errors before averaging, which gives more weight to larger errors.
RMSE provides an interpretable scale, as it is in the same units as the forecast
variable.

Problem 16:
a. True: Both MSE and RMSE square the errors before averaging, which gives
more weight to larger errors, thus penalizing them more heavily.
b. True: Both MSE and RMSE square the errors before averaging, which gives
more weight to larger errors, thus penalizing them more heavily.
c. True: MPE (Mean Percentage Error) provides information on bias; a consistent
positive MPE indicates the model predicts too low, while a consistent negative
MPE indicates it predicts too high.
d. False: While MAD (Mean Absolute Deviation) measures the average magnitude
of the errors, it does not express them in relation to the actual observation in a
percentage form like MAPE does.