Calculating first lag autocorrelation coefficient - repeated time values in sample - sample may not include multiple panels.

[Bas Boomen]

Hello,

I am fairly new to Stata and I am currently trying to find the AR(1) coefficient (which is to my knowledge the autocorrelation at lag 1) of my regression model.

My data looks as follows: in total I have 6000 unique lots over the time period 2009-2019

lot_identifier	Lot_price	Auction_year	auction_month	Auction_location	lot_size	ln_real_price	limited_edition_dummy
1	12560	2009	1	London	35	9.45	1
2	3400	2009	6	New York	56	8.24	0
3	2351	2010	1	London	12	4.245	1
4	453	2010	5	Dubai	34	7.32	1
5	6587	2011	7	Dubai	64	8.234	0
6	7809	2011	5	New York	12	4.34	0
7	4086	2011	8	Dubai	7	5.236	0
8	2354	2012	1	Dubai	23	2.45	1
9	2654	2012	9	London	24	2.234	0
10	3685	2013	10	Dubai	75	6.24	0
11	56966	2013	3	London	54	9.234	0
12	5373	2014	11	Dubai	53	1.34	1
13	9832	2015	3	New York	43	3.234	0
14	24609	2015	1	London	12	6.432	1
15	95028	2016	9	Dubai	67	7.235	0
16	3456	2017	12	New York	34	2.245	1
17	7795	2017	3	London	23	6.325	0
18	3264	2018	11	Dubai	64	1.325	1
19	9245	2019	3	Dubai	41	4.246	0

I peformed a regression like this to find the effects of the independent variables on the dependent variable 'ln_real_price:

reg ln_real_price i.auction_year i.auction_month i.auction_location lot_size limited_edition_dummy, robust baselevels

Now I want to find the autocorrelation of the first lag. First of all, I am not sure which method is best to use to find this coefficient. Secondly, when trying different methods Stata tells me to set a time variable. I want to set 'auction_year' as my time variable but when I do

tsset auction_year

stata gives me the error 'repeated time values in sample'.
Then if I try

tsset A auction_year

I get this

tsset A auction_year
panel variable: A (unbalanced)
time variable: auction_year, 2009 to 2019
delta: 1 unit

And if I then try for example

corrgram ln_real_price

or

wntestq ln_real_price, lags(1)

I get the error 'sample may not include multiple panels'.

Maybe I am using the totally wrong tests for finding the autocorrelation of the first lag of my regression model, maybe it is something else? Can somebody please help me to find the AR(1) coefficient?

Kind regards,

Bas van den Boomen

[Clyde Schechter]

-corrgram- is only usable with time series data. You do not have time series data: for most years you have several observations. So it is not even possible to define lags in this data. Look at the observation with lot_no = 6. The year is 2011. But there are three different observations for 2010: which one of them is supposed to be "the lag" of that observation? It is impossible to say. The very concept of lags and autocorrelation is inapplicable to this kind of data.

It looks, from your example data tableau, as if you really have panel data here, with three panels (New York, London, and Dubai). And maybe you want to do a separate -corrgram- for each of those? A separate command, with an appropriate -if- clause, for each of those cities would do the trick. If the real data has many panels, use a -foreach- loop.