advanced econometrics i
play

ADVANCED ECONOMETRICS I Theory (3/3) Instructor: Joaquim J. S. - PowerPoint PPT Presentation

ADVANCED ECONOMETRICS I Theory (3/3) Instructor: Joaquim J. S. Ramalho E.mail: jjsro@iscte-iul.pt Personal Website: http://home.iscte-iul.pt/~jjsro Office: D5.10 Course Website: https://jjsramalho.wixsite.com/advecoi Fnix:


  1. ADVANCED ECONOMETRICS I Theory (3/3) Instructor: Joaquim J. S. Ramalho E.mail: jjsro@iscte-iul.pt Personal Website: http://home.iscte-iul.pt/~jjsro Office: D5.10 Course Website: https://jjsramalho.wixsite.com/advecoi Fรฉnix: https://fenix.iscte-iul.pt/disciplinas/03089

  2. 3. Discrete Choice Models 3.2. Models for Ordered Choices Ordered choices: Values for the dependent variable: ๐‘ โˆˆ 0,1, โ€ฆ , ๐‘ โˆ’ 1 Latent model: โˆ— = ๐‘ฆ ๐‘— โ€ฒ ๐›พ + ๐‘ฃ ๐‘— ๐‘ ๐‘— โ–ช ๐‘ฆ ๐‘— cannot include an intercept Individual behaviour observed only by intervals: โˆ— โ‰ค ๐›ฟ 0 0 if ๐‘ ๐‘— โˆ— โ‰ค ๐›ฟ ๐‘› , ๐‘› if ๐›ฟ ๐‘›โˆ’1 < ๐‘ 1 โ‰ค ๐‘› โ‰ค ๐‘ โˆ’ 2 ๐‘ ๐‘— = เตž ๐‘— โˆ— > ๐›ฟ ๐‘โˆ’2 ๐‘ โˆ’ 1 if ๐‘ ๐‘— โ–ช Example: โˆ— is a latent measure of the health status โ€“ ๐‘ ๐‘— โ€“ ๐‘ ๐‘— is an observed health indicator: poor, satisfactory, good, excellent Assumption: the ๐›ฟ ๐‘˜ โ€™s are not known 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 2

  3. 3. Discrete Choice Models 3.2. Models for Ordered Choices Probabilities: Aim: โˆ— in a given interval โ–ช Modelling the probability of observing ๐‘ ๐‘— Each probability is based on the same ๐ป โˆ™ functions used with binary choices, being given by: โˆ— โ‰ค ๐›ฟ ๐‘› |๐‘ฆ ๐‘— ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘— = ๐‘„๐‘  ๐›ฟ ๐‘›โˆ’1 < ๐‘ ๐‘— โˆ— โ‰ค ๐›ฟ ๐‘› |๐‘ฆ ๐‘— โˆ’ ๐‘„๐‘  ๐‘ โˆ— < ๐›ฟ ๐‘›โˆ’1 |๐‘ฆ ๐‘— = ๐‘„๐‘  ๐‘ ๐‘— ๐‘— โ€ฒ ๐›พ + ๐‘ฃ ๐‘— โ‰ค ๐›ฟ ๐‘› |๐‘ฆ ๐‘— โˆ’ ๐‘„๐‘  ๐‘ฆ ๐‘— โ€ฒ ๐›พ + ๐‘ฃ ๐‘— < ๐›ฟ ๐‘›โˆ’1 |๐‘ฆ ๐‘— = ๐‘„๐‘  ๐‘ฆ ๐‘— โ€ฒ ๐›พ|๐‘ฆ ๐‘— โˆ’ ๐‘„๐‘  ๐‘ฃ ๐‘— < ๐›ฟ ๐‘›โˆ’1 โˆ’ ๐‘ฆ ๐‘— โ€ฒ ๐›พ|๐‘ฆ ๐‘— = ๐‘„๐‘  ๐‘ฃ ๐‘— โ‰ค ๐›ฟ ๐‘› โˆ’ ๐‘ฆ ๐‘— โ€ฒ ๐›พ โˆ’ ๐ป ๐›ฟ ๐‘›โˆ’1 โˆ’ ๐‘ฆ ๐‘— โ€ฒ ๐›พ = ๐ป ๐›ฟ ๐‘› โˆ’ ๐‘ฆ ๐‘— Hence, the general case is: โ€ฒ ๐›พ ๐ป ๐›ฟ 0 โˆ’ ๐‘ฆ ๐‘— if ๐‘› = 0 โ€ฒ ๐›พ โˆ’ ๐ป ๐›ฟ ๐‘›โˆ’1 โˆ’ ๐‘ฆ ๐‘— โ€ฒ ๐›พ if 1 โ‰ค ๐‘› โ‰ค ๐‘ โˆ’ 2 ๐ป ๐›ฟ ๐‘› โˆ’ ๐‘ฆ ๐‘— ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘— = เตž โ€ฒ ๐›พ 1 โˆ’ ๐ป ๐›ฟ ๐‘โˆ’2 โˆ’ ๐‘ฆ ๐‘— if ๐‘› = ๐‘ โˆ’ 1 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 3

  4. 3. Discrete Choice Models 3.2. Models for Ordered Choices Estimation: Parameters to be estimated: โ–ช ๐›พ โ–ช ๐›ฟ 0 , โ€ฆ , ๐›ฟ ๐‘โˆ’2 Estimation method: โ–ช Maximum likelihood Most common models: โ–ช Ordered logit โ–ช Ordered probit Stata ologit Y ๐‘Œ 1 โ€ฆ ๐‘Œ ๐‘™ oprobit Y ๐‘Œ 1 โ€ฆ ๐‘Œ ๐‘™ 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 4

  5. 3. Discrete Choice Models 3.2. Models for Ordered Choices Partial effects: Each ๐‘Œ ๐‘˜ affects ๐‘ probabilities: โˆ†๐‘Œ ๐‘˜ = 1 โŸน โˆ†๐‘„๐‘  ๐‘ = ๐‘› ๐‘Œ โ€ฒ ๐›พ โˆ’๐›พ ๐‘˜ ๐‘• ๐›ฟ 0 โˆ’ ๐‘ฆ ๐‘— if ๐‘› = 0 โ€ฒ ๐›พ โˆ’ ๐‘• ๐›ฟ ๐‘›โˆ’1 โˆ’ ๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐›พ ๐‘˜ ๐‘• ๐›ฟ ๐‘› โˆ’ ๐‘ฆ ๐‘— if 1 โ‰ค ๐‘› โ‰ค ๐‘ โˆ’ 2 = เตž โ€ฒ ๐›พ ๐›พ ๐‘˜ ๐‘• ๐›ฟ ๐‘โˆ’2 โˆ’ ๐‘ฆ ๐‘— if ๐‘› = ๐‘ โˆ’ 1 The sign of ๐›พ ๐‘˜ is informative about the direction of โˆ†๐‘„๐‘  ๐‘ = 0 ๐‘Œ and โˆ†๐‘„๐‘  ๐‘ = ๐‘ โˆ’ 1 ๐‘Œ but not of the changes in the remaining probabilities 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 5

  6. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Multinomial choices: Values for the dependent variable: ๐‘ โˆˆ 0,1, โ€ฆ , ๐‘ โˆ’ 1 Latent model: โ–ช Each individual has a given utility associated with each alternative: โ€ฒ ๐›พ + ๐‘ฃ ๐‘—๐‘› ๐‘‰ ๐‘—๐‘› = ๐‘ฆ ๐‘—๐‘› โ–ช The selected alternative is the one that maximizes utility: ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘— = ๐‘„๐‘  ๐‘‰ ๐‘—๐‘› = ๐‘›๐‘๐‘ฆ ๐‘‰ ๐‘—1 , โ€ฆ ๐‘‰ ๐‘—๐‘ |๐‘ฆ ๐‘— Main models: โ–ช Multinomial Logit: ๐‘‰ ๐‘—๐‘› ~ ๐ป๐‘ฃ๐‘›๐‘๐‘“๐‘š and ๐‘‰ ๐‘—๐‘› independent โˆ€๐‘› โ–ช Multinomial Probit: ๐‘‰ ๐‘—๐‘› ~ ๐‘‚๐‘๐‘ ๐‘›๐‘๐‘š โ–ช Nested Logit โ–ช Random Parameters Logit 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 6

  7. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Explanatory variables: ๐‘ฆ ๐‘—๐‘› may include: โ–ช ๐‘ฆ ๐‘—๐‘› : variables that are different across individuals and alternatives โ–ช ๐‘ฆ ๐‘› : variables that differ across alternatives but not individuals โ–ช ๐‘ฆ ๐‘— : variables that differ across individuals but not alternatives Example: โ–ช ๐‘ ๐‘— - selected means of transport to go to work โ–ช ๐‘ฆ ๐‘—๐‘› - time that each individual ๐‘— takes in going to work when using transport ๐‘› โ–ช ๐‘ฆ ๐‘› - price of transport ๐‘› โ–ช ๐‘ฆ ๐‘— - age of individual ๐‘— 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 7

  8. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Multinomial logit: โ€ฒ ๐›พ+๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘› ๐‘“ ๐‘ฆ ๐‘—๐‘› โ€ฒ ๐›พ + ๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘› = ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘—๐‘› = ๐ป ๐‘› ๐‘ฆ ๐‘—๐‘› โ€ฒ ๐›พ+๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘˜ ๐‘โˆ’1 ๐‘“ ๐‘ฆ ๐‘—๐‘˜ ฯƒ ๐‘˜=0 ๐›พ ๐‘› has to be normalized, that is for one alternative (base outcome) its value is set to zero ๐›พ cannot include a constant term Independence of Irrelevant Alternatives (IIA) โ€“ the odds ratio between two alternatives does not depend on the remaining alternatives: โ€ฒ ๐›พ+๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘› = ๐‘“ ๐‘ฆ ๐‘—๐‘› ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘—๐‘› โ€ฒ ๐›พ+๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘š ๐‘„๐‘  ๐‘ ๐‘— = ๐‘š|๐‘ฆ ๐‘—๐‘› ๐‘“ ๐‘ฆ ๐‘—๐‘š 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 8

  9. 3. Discrete Choice Models 3.3. Models for Multinomial Choices When all explanatory variables are of the type ๐‘ฆ ๐‘—๐‘› and ๐‘ฆ ๐‘› , the choice between alternatives ๐‘› and ๐‘š is fully explained by diferences in the alternative characteristics: โ€ฒ ๐›พ = ๐‘“ ๐‘ฆ ๐‘—๐‘› ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘—๐‘› โ€ฒ โˆ’๐‘ฆ ๐‘—๐‘š โ€ฒ ๐›พ โ€ฒ ๐›พ = ๐‘“ ๐‘ฆ ๐‘—๐‘› ๐‘„๐‘  ๐‘ ๐‘— = ๐‘š|๐‘ฆ ๐‘—๐‘š ๐‘“ ๐‘ฆ ๐‘—๐‘š โ–ช Is this case, the model is often called โ€˜ conditional logit โ€™ Stata asclogit Y ๐‘Œ 1๐‘› โ€ฆ, case( id ) alternatives( varname ) casevars( ๐‘Œ ๐‘— โ€ฆ) basealternative( name ) When all explanatory variables are of the type ๐‘ฆ ๐‘— , the choice between alternatives ๐‘› e ๐‘š is fully explained by diferences between ๐›พ ๐‘› e ๐›พ ๐‘š : โ€ฒ ๐›พ ๐‘› = ๐‘“ ๐‘ฆ ๐‘— ๐‘„๐‘  ๐‘ ๐‘— = ๐‘›|๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘› โˆ’๐›พ ๐‘š โ€ฒ ๐›พ ๐‘š = ๐‘“ ๐‘ฆ ๐‘— ๐‘„๐‘  ๐‘ ๐‘— = ๐‘š|๐‘ฆ ๐‘— ๐‘“ ๐‘ฆ ๐‘— Stata mlogit Y ๐‘Œ 1 โ€ฆ ๐‘Œ ๐‘™ , baseoutcome( 0 ) 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 9

  10. าง 3. Discrete Choice Models 3.3. Models for Multinomial Choices Estimation: โ–ช Maximum likelihood based on the following log-likelihood function: ๐‘‚ โ€ฒ ๐›พ + ๐‘ฆ ๐‘— โ€ฒ ๐›พ ๐‘› ๐‘€๐‘€ = เท ๐‘’ ๐‘—๐‘› ๐‘š๐‘๐‘• ๐ป ๐‘› ๐‘ฆ ๐‘—๐‘› ๐‘—=1 โ–ช ๐‘’ ๐‘—๐‘› = 1 if individual ๐‘— chooses alternative ๐‘› Partial effects: โ–ช โˆ†๐‘Œ ๐‘—๐‘˜ = 1 โŸน โ€“ โˆ†๐‘„๐‘  ๐‘ ๐‘— = ๐‘› ๐‘Œ = ๐›พ ๐‘˜ ๐ป ๐‘› โˆ™ ๐‘’ ๐‘—๐‘› โˆ’ ๐ป ๐‘› โˆ™ โ€“ ๐›พ ๐‘˜ gives the sign of the partial effect โ–ช โˆ†๐‘Œ ๐‘— = 1 โŸน ๐‘โˆ’1 ๐›พ ๐‘› ๐ป ๐‘› โˆ™ โ€“ โˆ†๐‘„๐‘  ๐‘ ๐›พ = ฯƒ ๐‘›=1 ๐‘— = ๐‘› ๐‘Œ = ๐ป ๐‘› โˆ™ ๐›พ ๐‘˜ โˆ’ ๐›พ , onde าง โ€“ ๐›พ ๐‘˜ gives the sign of the partial effect relative to the base alternative, not the sign of the overall effect 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 10

  11. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Testing IIA โ–ช Hausman test comparing: โ€“ Full multinomial logit model โ€“ Multinomial logit model excluding one or more alternatives โ–ช If multinomial logit is the correct model, then both models produce consistent estimators (null hypothesis) โ–ช If multinomial logit is not the correct model, then the results generated by both models will be different (alternative hypothesis) Stata mlogit Y ๐‘Œ 1 โ€ฆ ๐‘Œ ๐‘™ , baseoutcome( 0 ) (ou asclogit โ€ฆ) estimates store Mod1 mlogit Y ๐‘Œ 1 โ€ฆ ๐‘Œ ๐‘™ if Y != 3 , baseoutcome( 0 ) (ou asclogit โ€ฆ) estimates store Mod2 hausman Mod1 Mod2 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 11

  12. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Multinomial probit: Not affected by the IIA property Very complex, requiring the computation of ๐‘ โˆ’ 1 integrals The version implemented in Stata assumes independent errors, which eliminates the only advantage of multinomial probit over multinomial logit Stata mprobit Y ๐‘Œ 1 โ€ฆ ๐‘Œ ๐‘™ , baseoutcome( 0 ) 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 12

  13. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Nested logit: Not affected by the IIA property, grouping the choices in several sets in such a way that: โ–ช Within each group, alternatives may be correlated โ–ช Between groups, alternatives are independent Results from a sequential decision process โ€“ example for a two-level process: Financing Bank debt Own funds Stock market Bank 1 Bank N Lisbon Frankfurt Stata โ–ช Level 1 โ€“ defining J groups, nlogit โ€ฆ โ–ช Level 2 โ€“ defining ๐‘ ๐‘˜ choices in each group 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 13

  14. 3. Discrete Choice Models 3.3. Models for Multinomial Choices Random parameters logit: Latent model: โ€ฒ ๐›พ ๐‘— + ๐‘ฃ ๐‘—๐‘› ๐‘‰ ๐‘—๐‘› = ๐‘ฆ ๐‘—๐‘› Most common assumption: ๐›พ ๐‘— ~ ๐‘‚ ๐›พ, ฮฃ ๐›พ Not affected by the IIA property If ฮฃ ๐›พ = 0 , it reduces to the Multinomial Logit model; hence, comparing the two models allows the IIA property to be tested 2020/2021 Joaquim J.S. Ramalho Advanced Econometrics I 14

Recommend


More recommend