Hi,
I am trying to estimate demand, and subsequently compute elasticities, in the cigarette product market in Chicago using scanner data from UChicago's dataset on Dominick's Finer Foods (DFF) chain. My period of analysis is 1994-1997 and my nesting structure is a one-level nest with Class A brands (with a per pack price of >$2) as one group and Class B brands (with a per pack price of <$2) as another group, and an outside option (generally normalised to zero). My market definition is on a store-week level and market size is the weekly customer count, defined in the DFF manual as "The customer count data refers to the number of customers visiting the store and purchasing something." I used the following commands:
egen combination=group(store week), label
xtset UPC combination
gen str20 nest2="."
replace nest2 = "Class A cigarettes" if final_brand=="Benson & Hedges" | final_brand=="Kool" | final_brand=="Marlboro" | final_brand=="Virginia" | final_brand=="Winston"
replace nest2 = "Class B cigarettes" if final_brand=="Generic" | final_brand=="Sub-generic"
encode nest2, gen(nest2_num)
mergersim init, nests(nest2_num) price(price) quantity(move) marketsize(weekly_ccount)
xtreg M_ls price M_lsjg _size length1_characteristic length2_characteristic premium_characteristic menthol_characteristics slim_characteristics i.store i.week
I also plan on adding brand-fixed effects, which I have omitted from posting on this forum to keep things simple to understand. By common understanding and literature, I understand that price and M_lsjg i.e. price and within-group market shares are endogenous. As suggested by Berry [1994] valid instruments are cost variables and the characteristics of competing brands in this group. In practice, when doing nested logit, researchers often use the number and mean characteristics of other products as instruments, especially, the number and mean characteristics of other products in the same group and subgroup are added to instrument within shares (see Verboven, 1996; Bresnahan, Stern, and Trajtenberg, 1997).
Because in my dataset, cost variable has no variation, I choose to only use product characteristics as my instruments (the famous BLP instruments). I understand conceptually what my instruments should be but I do not know how to construct a variable for characteristics for "other products" or "other brands" when I am not really running the regressions on one brand j to be able to instrument using characteristics of all other brands (excluding j).
Can someone please help me with instrument construction based on my regressions?
I am trying to estimate demand, and subsequently compute elasticities, in the cigarette product market in Chicago using scanner data from UChicago's dataset on Dominick's Finer Foods (DFF) chain. My period of analysis is 1994-1997 and my nesting structure is a one-level nest with Class A brands (with a per pack price of >$2) as one group and Class B brands (with a per pack price of <$2) as another group, and an outside option (generally normalised to zero). My market definition is on a store-week level and market size is the weekly customer count, defined in the DFF manual as "The customer count data refers to the number of customers visiting the store and purchasing something." I used the following commands:
egen combination=group(store week), label
xtset UPC combination
gen str20 nest2="."
replace nest2 = "Class A cigarettes" if final_brand=="Benson & Hedges" | final_brand=="Kool" | final_brand=="Marlboro" | final_brand=="Virginia" | final_brand=="Winston"
replace nest2 = "Class B cigarettes" if final_brand=="Generic" | final_brand=="Sub-generic"
encode nest2, gen(nest2_num)
mergersim init, nests(nest2_num) price(price) quantity(move) marketsize(weekly_ccount)
xtreg M_ls price M_lsjg _size length1_characteristic length2_characteristic premium_characteristic menthol_characteristics slim_characteristics i.store i.week
I also plan on adding brand-fixed effects, which I have omitted from posting on this forum to keep things simple to understand. By common understanding and literature, I understand that price and M_lsjg i.e. price and within-group market shares are endogenous. As suggested by Berry [1994] valid instruments are cost variables and the characteristics of competing brands in this group. In practice, when doing nested logit, researchers often use the number and mean characteristics of other products as instruments, especially, the number and mean characteristics of other products in the same group and subgroup are added to instrument within shares (see Verboven, 1996; Bresnahan, Stern, and Trajtenberg, 1997).
Because in my dataset, cost variable has no variation, I choose to only use product characteristics as my instruments (the famous BLP instruments). I understand conceptually what my instruments should be but I do not know how to construct a variable for characteristics for "other products" or "other brands" when I am not really running the regressions on one brand j to be able to instrument using characteristics of all other brands (excluding j).
Can someone please help me with instrument construction based on my regressions?
