DETECTION OF OUTLIERS IN THE CANADIAN CONSUMER PRICE INDEX (CPI)

1

• DETECTION OF OUTLIERS IN THE CANADIAN
  CONSUMER PRICE INDEX (CPI)
• ABDELNASSER SAÏDI AND SUSANA RUBIN BLEUER
• STATISTICS CANADA

2
What is the CPI ?

• Is a weighted arithmetic average of price indexes
  over all basic classes. The weights are obtained
  from the survey of households spending.
• the price index at the basic class is in turn a
  weighted sum of the micro indexes over all the
  items in the class.
• The micro index is a geometric mean of the price
  relatives of n outlets surveyed.

3
CURRENT EVALUATION PROCEDURES

• Large amount of resources used to correct a small
  number of errors.
• verifications made manually
• Verification of all records with status
  special.
• Verification of all records with a relative price
  increased or decreased by 15.
• Editing procedures related to missing values,
  missing outlets and quality changes (see
  Allard-Saulnier and Beyrouti (2003)).

4
PROPOSED EVALUATION PROCEDURES

• Performed on an editing group
• All observations pertaining to a particular item
• and in a particular geographical area ( 4
  possible levels)
• and having the same status code if the percentage
  of SPs gt 15.
• status code  SPs 
• ( regular ? special or special?
  regular).

5
PROPOSED EVALUATION PROCEDURES

• Methods considered
• ILO Tukey and variants,
• MAD (Median Absolute Deviation),
• Quartile method and HB (Hidiroglou-Berthelot)
• RFM (Resistant Fence Method)
• Detection method with higher breakdown point
  than Tukeys method and taking into account the
  skewness of the distribution and the small size
  of the CPI sample are favoured
• Quartile method with asymmetric fences applied to
  the log of the price relatives was recommended

6
INFLUENCE IN THE BASIC CLASS

• Contribution of each outlier on the basic class
  index
• Analysis of the percentage of contribution
• How to define a cut-off to identify influent
  observations ?