missing prices for certain car models: tips for predicting missing values

[Michael Duarte Goncalves]

Hi Statalist Community,

I need your help, please.

I have to predict the prices of certain models for which I don't have the prices. I have two databases, one with registrations, the other with prices. After matching, this is what I get. I'm particularly interested in -merged == 3- and -merged == 1-. For vehicles in -merged == 3-, all the prices are present.

. // merge process:
. merge m:1 id using `car_prices', generate(merged)
(variable model was str39, now str60 to accommodate using data's values)
(variable fuel was str1, now str4 to accommodate using data's values)

    Result      Number of obs
    -----------------------------------------
    Not matched      2,661,329
        from master   2,659,732 (merged==1)
        from using        1,597 (merged==2)

    Matched         1,340,682 (merged==3)
    -----------------------------------------

What I'm trying to predict are the missing prices of cars from -merged == 1-. Roughly speaking, here is a small extract from my database (one line to get an idea and avoid overloading):

* Example generated by -dataex-. For more info, type help dataex
clear
input str23 brand str60 model str4 fuel long engine_cap double power_kw int co2 float fiscal_power str4 ev_type double price
‘ALFA ROMEO’ ‘146’ ‘1’ 1929 0 . 13.03 ‘’ .
end

Two of these variables are good predictors of price: -engine_cap- and -power_kw-: if these two variables increase, the price will also increase, on average.

But, in my price database, I also have atypical models (HEV, FCEV, BEV, etc.). As a result, for these models, the cubic capacity is 0 (variable -engine_cap-), since they don't burn air like petrol- or diesel-fuelled cars. So I was thinking of using fixed effects to control this.

. tab ev_type

         ev | Freq.     Percent      Cum.
------------+-----------------------------------
        BEV | 13,589       6.00     6.00
       FCEV |      2       0.00     6.01
        HEV |201,511      89.05    95.05
       PHEV | 10,873       4.80    99.86
       REEV |    320       0.14   100.00
------------+-----------------------------------
      Total |226,295     100.00

• This is what I do, and this is where I need your help, please. Do you think the following makes sense?

keep if merged == 1 | merged == 3


* Replace the gaps in the ev variable :
replace ev_type = ‘ combustion ’ if missing(ev_type) & (merged == 3)
encode ev_type, gen(ev_type_encode)
*
* Creation of a price log
gen ln_p = ln(price)
egen brand_model = group(brand model)


gen date = date(FEC_MATRICULA, ‘ DMY ’)
format date %td
*
*
* Step 1: Estimate the model and record the fixed effects
reghdfe ln_p engine_cap power_kw i.ev_type_encode, absorb(brand_model, savefe)

* Step 2: Generate predicted logarithmic prices for observed prices
predict yhat, xb
gen ln_p_hat = yhat + __hdfe1__

* Step 3: Calculate the average fixed effect for each brand model
egen efecto_fijo = mean(__hdfe1__), by(brand_model)

* Steps 4 and 5: Generate the predicted logarithmic prices for all observations
gen fixed_effect = __hdfe1__
replace fixed_effect = efecto_fijo if missing(fixed_effect)
gen ln_p_hat_all = yhat + fixed_effect


* Step 6: Generate the scatter plot
two-way scatter plot ln_p ln_p_hat_all if !missing(ln_p), ///
    xline(0) yline(0) ///
    title(‘Scatter plot of observed versus predicted logarithmic prices’) ///
    xtitle(‘Predicted logarithmic prices’) ///
    ytitle(‘Observed log prices’) ///
    legend(off)

• Does anyone know why this sentence appears when I used my regression above?

note: 3.ev_type_encode omitted because of collinearity

reghdfe ln_p engine_cap power_kw i.ev_type_encode, absorb(brand_model, savefe)

Thank you in advance for your help!

Michael