Time Series forcasting

Question 1

What is Time Series forecasting

Answer 1

refers to what is observed over time, generally speaking, forecasting aims to estimate how the sequence of observation will continue over time

Question 2

what are the types of forecasting

Answer 2

• the Explanatory model, where the variables actually capture the reason for the change
  Electricity Demand = f (current temperature, strength of economy, population,time of day, day of week,
  error).
• Time Series model where the prediction of the future is based on past values
  Electricity Demand EDt+1= f (EDt, EDt−1, EDt−2, EDt−3,…, error)
• mixed-model, a mixture of both

Question 3

what are the main steps of forecasting

Answer 3

1- initially define the problem what to forecast?
2- father the information 
3- Explore the data, trends seasonality 
4- choose the adequate fitting model 
5- evaluate your forecasting model

Question 4

why we perform time series decomposition

Answer 4

Decomposition offers a useful abstract model for the data

Question 5

what are the main time series components

Answer 5

the Trend, the long term increase or the decrease of the data
Seasonality, where the time series is affected by the seasonal factor
cycles similar to trend but lacks a fixed frequency

Question 6

what are the two types of decomposition

Answer 6

we have additive decomposition yt= St + Tt + Rt

and multiplicative one yt= St × Tt × Rt.

Question 7

what are the types of adjustment we need to add for the decomposition

Answer 7

this include calendar adjustment for instance sum of daily sales instead of monthly one as Not all month days are actually worked.

Question 8

how to estimate the trend Cycle

Answer 8

Use the moving average
● m=2k+1, k is number of neighbours
● Every value is replaced by the average of itself and its k nearest neighbourhood
● Results in a smooth trend-cycle component
● Ex: K=2, m=5, the result is called 5-MA
● Averaging is applied using a sliding window

Question 9

why use a Moving average of Moving average is

Answer 9

since odd-order moving averages is symmetric, while even order ones are not

Question 10

How to perform additive decomposition

Answer 10

1- compute the trend cycle Tt using MA
2- calculate the detrended series yt− Tt
3- compute the seasonal component for each season, it actually corresponds to the detrended values for that season in all the data
4 - compute the remaining

Question 11

suggest a simple forcasting

Answer 11

Mean method where all the future data is equal to the average of the history
Naive method, the future is based on the last observed one
Season naive each forecast is equal to the last observed values in the sam season

Question 12

how can we forecast using decomposition

Answer 12

Seasonality adjusted component A = T + R where R is the remainder component
forecast both the seasonally adjusted one and A
The seasonal component is simply repeated

Question 13

what are the stationary time series

Answer 13

sort of consistency in the data, where the distribution depends only in difference in time rather than location in time

Question 14

why it is beneficial to transform time series into a stationary one

Answer 14

This is because the statistical parameters, mean, variance does not vary over the time windows, hence we obtain a better forecast

Question 15

how to convert time series to stationary

Answer 15

perform a simple difference y i = y i − y i−1 and combine them with a log scale of one
We can also use the second-order differencing, and the obtained data model is y i+1 = y i+ c + e i+1
where e i+1 is the white noise with zero mean

Question 16

Explain the autocorrelation as a prediction tool

Answer 16

Autocorrelations represent the correlations between lags in time series

Question 17

what is autoregressive model

Answer 17

Auto regressive model performs a forecasting based on the previous values

Question 18

what is autoregressive model

Answer 18

Autoregressive model performs forecasting based on the previous window of size p, we call it AR(p) since P reflects the values we rely on

Question 19

what is the downside of AR

Answer 19

it can not always capture all the variations while moving average can capture both hence they are used together , in other words
● AR forcasr a series base solely on the past values in the series
while MA forecast series base solely on the past errors in the series called error lags

Question 20

Explain the Arima model

Answer 20

mixture or AR and MA