Time Series Forecasting Quiz

Question 1: Model Identification

Examine the ACF and PACF plots below:

Based on these ACF and PACF plots, which time series model is most appropriate for this data?

AR(1) model, as the PACF cuts off after lag 1 and the ACF shows a gradual decay MA(1) model, as the ACF cuts off after lag 1 and the PACF shows a gradual decay ARMA(1,1) model, as both ACF and PACF show gradual decay patterns White noise process, as neither ACF nor PACF show significant values at any lag

Question 2: Model Identification

Examine the ACF and PACF plots below:

Based on these ACF and PACF plots, which time series model is most appropriate for this data?

AR(2) model, as the PACF cuts off after lag 2 and the ACF shows a gradual decay MA(2) model, as the ACF cuts off after lag 2 and the PACF shows a gradual decay ARMA(2,2) model, as both ACF and PACF show significant values at multiple lags White noise process, as most ACF values fall within the confidence intervals

Question 3: Seasonality Detection

The chart below shows monthly sales data for a retail company over a 3-year period:

What can be concluded about the seasonality in this retail sales data?

There is no clear seasonal pattern in this data, as sales fluctuate randomly throughout the period There is annual seasonality with peaks in December and troughs in mid-summer months There is quarterly seasonality with peaks occurring every three months in the data There is semi-annual seasonality with two major peaks each year in the data

Question 4: ARIMA Model Results

The following table shows the results of fitting several ARIMA models to a time series dataset:

Model	AIC	BIC	RMSE	MAE	Residual Q-test p-value
ARIMA(1,1,0)	326.45	334.21	4.32	3.45	0.03
ARIMA(0,1,1)	322.18	329.94	4.25	3.38	0.08
ARIMA(1,1,1)	318.72	329.86	4.15	3.30	0.21
ARIMA(2,1,0)	324.56	335.70	4.28	3.41	0.07
ARIMA(0,1,2)	320.34	331.48	4.20	3.35	0.15

Based on the model comparison results, which ARIMA model should be selected as the best for this time series?

ARIMA(1,1,0) because it has the simplest structure with fewest parameters ARIMA(1,1,1) because it has the lowest AIC, reasonable BIC, and non-significant residual Q-test ARIMA(0,1,2) because it balances complexity with relatively good performance metrics ARIMA(0,1,1) because it has better BIC than ARIMA(1,1,1) and acceptable error metrics

Question 5: Forecast Accuracy Comparison

The table below compares the forecast accuracy of different time series models for quarterly revenue predictions:

Model	RMSE	MAE	MAPE	Training Time (s)	Number of Parameters
ARIMA(2,1,1)	145.32	112.45	8.7%	1.2	3
Prophet	132.18	105.76	7.9%	3.5	NA
Holt-Winters	128.43	102.31	7.5%	0.9	3
SARIMA(1,1,1)(1,0,1)4	126.85	100.24	7.2%	2.1	5

Based on this comparison, which model would you select if the primary concern is forecast accuracy?

ARIMA(2,1,1) because it's simpler and requires less training time than other models Prophet because it balances good accuracy with flexibility in modeling trend components Holt-Winters because it offers strong accuracy with minimal training time SARIMA(1,1,1)(1,0,1)4 because it has the lowest RMSE, MAE, and MAPE values

Question 6: Time Series Decomposition

Below is a classical decomposition of monthly hotel bookings data:

What is the most accurate description of this time series based on its decomposition?

The data shows a strong upward trend with minimal seasonality The data has a downward trend with a clear 6-month seasonal pattern The data has a slight downward trend with a clear annual seasonal pattern The data is predominantly random with no clear trend or seasonal components

Question 7: Prophet Model Components

Below are the component plots from a Facebook Prophet model for an e-commerce company's daily sales data:

Based on the Prophet component plots, what is the most accurate business interpretation?

Sales are highest on weekends, peak during summer months, and show a clear downward trend over time Sales have an upward trend, are highest on Tuesdays and Saturdays, and peak in November-December Sales show a declining trend, are highest mid-week, and peak during summer with significant holiday effects Sales are expected to increase in the future, with major peaks during holiday periods and weekends

Question 8: Stationarity Assessment

Below are two time series plots, along with their respective Augmented Dickey-Fuller (ADF) test results:

Based on the time series plots and ADF test results, which statement is correct?

Both Series A and Series B are stationary as they both have constant variance over time Series A is non-stationary with a clear trend, while Series B is stationary, as confirmed by their respective ADF test results Both Series A and Series B are non-stationary as indicated by their ADF test statistics below critical values Series A becomes stationary after second differencing while Series B requires first differencing

Question 9: SARIMA Model Selection

Below is a time series of monthly airline passenger counts along with the ACF and PACF of the differenced series:

Based on the time series characteristics and model comparison, which SARIMA model specification is most appropriate for this airline passenger data?

ARIMA(1,1,1)(0,0,0) because it has the simplest structure requiring fewer parameters ARIMA(1,1,1)(1,0,1)12 because it accounts for both regular and seasonal patterns in the data ARIMA(0,1,1)(0,1,1)12 because it has the lowest AIC/BIC and accounts for seasonal differencing A different model is needed as none of these properly account for the quarterly pattern seen in the data

Question 10: Residual Analysis

The following plots show residual diagnostics for an ARIMA model fit to quarterly GDP growth:

Based on these residual diagnostic plots, what can you conclude about the ARIMA model fit?

The model is well-specified as residuals appear random with no autocorrelation and follow a normal distribution The model is inadequate because the residuals show clear patterns over time and significant autocorrelation The model has issues with non-normality as shown by significant deviation in the QQ plot and histogram The model needs seasonal components as the ACF shows spikes at seasonal lags not captured by the model

Question 11: Forecasting Method Comparison

Below is a comparison of forecast results from Holt-Winters and SARIMA models for quarterly sales data:

Based on the forecast comparison chart and error metrics, which statement is most accurate about the two forecasting methods?

Holt-Winters captures the seasonal pattern better but has higher error metrics than SARIMA SARIMA provides more accurate forecasts as indicated by lower MAPE and RMSE values Holt-Winters has wider confidence intervals, making it more reliable for long-term forecasting Both methods perform similarly, but SARIMA requires more complex parameter specification

Question 12: Time Series Transformations

The plots below show a time series before and after different transformations:

Based on the transformations and ADF test results, which statement is correct about achieving stationarity?

The original series is already stationary since it has a clear trend pattern The log transformation successfully achieved stationarity as shown by the p-value below 0.05 First differencing is the only transformation that successfully achieved stationarity None of the transformations achieved stationarity; second differencing would be required

Question 13: Prophet Components Analysis

The following table shows the breakdown of Prophet model components for daily website traffic:

Component	Average Effect	Min Effect	Max Effect	% of Variation
Trend	+42.3%	+12.5%	+68.4%	38.2%
Weekly Seasonality	±15.8%	-22.4%	+28.7%	34.5%
Yearly Seasonality	±8.3%	-12.6%	+15.2%	21.3%
Holiday Effects	±12.1%	-18.3%	+45.9%	6.0%

Based on the Prophet component analysis, what insight about the website traffic is most strongly supported?

The website traffic is primarily influenced by holiday effects that cause dramatic spikes The dominant pattern is strong upward growth with significant weekly variations Yearly seasonality is the most important factor to consider when planning website capacity The website traffic is highly unpredictable as no single component explains more than 40% of variation

Question 14: Forecast Evaluation

The chart below shows 6-month rolling forecast accuracy metrics for a demand forecasting model:

Based on the forecast accuracy chart, what can be concluded about the forecasting model's performance?

The model's accuracy deteriorated after the model change in July 2023 The model change in July 2023 significantly improved forecast accuracy, reducing both MAPE and MAE The new product launch had a more disruptive effect on forecast accuracy than the promotion period There is no clear pattern in forecast accuracy; it varies randomly throughout the time period

Question 15: ARMA Model Selection

The following table summarizes different ARMA model fits for a stationary time series of monthly interest rates:

Model	AR Order (p)	MA Order (q)	Log-Likelihood	AIC	BIC	Ljung-Box (p-value)
ARMA(1,0)	1	0	-132.45	268.90	274.16	0.06
ARMA(0,1)	0	1	-136.78	277.56	282.82	0.03
ARMA(1,1)	1	1	-129.32	264.64	272.53	0.21
ARMA(2,0)	2	0	-130.16	266.32	274.21	0.18
ARMA(0,2)	0	2	-134.89	275.78	283.67	0.07
ARMA(2,1)	2	1	-128.94	265.88	276.40	0.32
ARMA(1,2)	1	2	-129.08	266.16	276.68	0.28

Based on the model comparison results, which ARMA model would you select as the most appropriate?

ARMA(1,0) because it has the simplest structure while maintaining reasonable fit statistics ARMA(1,1) because it has the lowest AIC value and a non-significant Ljung-Box test ARMA(2,1) because it has the highest log-likelihood and best Ljung-Box p-value ARMA(2,0) because it balances model complexity with good fit statistics and diagnostic tests

Question 16: Holt-Winters Parameter Analysis

The table below shows the parameter estimates for a multiplicative Holt-Winters model applied to quarterly product sales data:

Parameter	Estimate	Lower 95% CI	Upper 95% CI	Description
α (alpha)	0.82	0.65	0.94	Level smoothing
β (beta)	0.05	0.01	0.18	Trend smoothing
γ (gamma)	0.75	0.58	0.89	Seasonal smoothing
Seasonal Indices
Q1	0.84	0.80	0.88	Q1 factor
Q2	1.12	1.08	1.16	Q2 factor
Q3	1.26	1.22	1.30	Q3 factor
Q4	0.78	0.74	0.82	Q4 factor

Based on the Holt-Winters parameter estimates, which interpretation is most accurate?

The model places heavy emphasis on recent observations for level with strong seasonality but minimal trend The model suggests sales are highest in Q4 (holiday season) with only slight variations across quarters The high β value indicates a strong trend component that dominates the seasonal effects in the model The model indicates inconsistent seasonality as shown by wide confidence intervals for the seasonal factors

Question 17: Seasonality Patterns

Below is a seasonal subseries plot for monthly hotel booking data:

Based on the seasonal subseries plot, which statement best describes the seasonal pattern in hotel bookings?

Bookings peak in late spring (May) and gradually decline throughout the rest of the year There are two distinct peak seasons: summer (June-August) and winter (December-January) Bookings peak in early summer (June-July) and early fall (September-October), with lowest values in winter The highest bookings occur in summer (June-August), with a secondary peak in April-May

Question 18: Autocorrelation Analysis

Below is the autocorrelation function (ACF) plot for daily stock returns of a technology company:

Based on the ACF plot and test results, what can be concluded about this stock returns time series?

The stock returns show significant positive autocorrelation at multiple lags, violating market efficiency The stock returns appear to follow a random walk pattern with no significant autocorrelation The stock returns show a significant seasonal pattern with peaks every 10 lags The stock returns show significant negative autocorrelation suggesting mean reversion tendencies

Question 19: Error Metrics Comparison

The following chart shows forecast errors (MAPE) for different time series methods across multiple forecast horizons:

Based on the error comparison chart, which statement is most accurate?

ARIMA performs better than all other methods for short-term forecasts but degrades for longer horizons Prophet consistently outperforms other methods across all forecast horizons, especially for long-term forecasts Holt-Winters shows more stable performance than SARIMA across different forecast horizons All methods show similar performance for short-term forecasts but diverge significantly at longer horizons

Question 20: Prophet Changepoint Analysis

The plot below shows a Prophet model fit with changepoints for monthly revenue data:

Based on the Prophet changepoint analysis, what is the most accurate interpretation of the revenue trend?

Revenue shows continuous growth with increasing rates of change at each detected changepoint Revenue has been consistently declining, with each changepoint showing a steeper downward trend Revenue experienced an initial decline followed by stability, with the most significant changepoint in May 2021 Revenue shows seasonal patterns with changepoints occurring at similar times each year

Time Series Forecasting Models Quiz

Student Information