Tracking Asking Prices


#1

Do any sites track properties from their Initial Asking Price -> Price Drops/Increases -> Sale Price as per the PPR?

I think this would be useful to see what properties are actually selling for but I cannot seem to find any resource. Anyone know of a resource that does this already?

Failing that looking it MyHome it seems one could scrape the data on the “Recently Added” page to populate a DB of new properties as they appear on the site and index them with the MyHome property id. One could then use that unique identifier to subsequently match any new entries on the “Price Changes” page to track any movement on the asking price. However the “Sold Properties” page does not make use of the same identifier and one would need to create a regex to match the address string with entries stored in the DB.


#2

I’ve been doing this intermittently for a couple of years – search for threads with PPR in the title. A fundamental problem is the difficulty of matching addresses between the PPR and myhome, because of data quality plus the basic ambiguity of Irish addresses. Mantissa recently suggested a new matching approach based on Google geocoding, and I’ll be pursuing that in the next while.


#3

So I was thinking that using the Selenium Webdriver tool (myhome.ie/recent) on MyHome.

Looking at the source of the page I see that associated with each entry is a unique identifier (e.g. myhome.ie/residential/brochu … 22/2878663).

One could also track the “Price Changes” page and use the unique identifier to index back into ones local DB to match an existing entry and then add in one or more price changes.

The tricky part (as you correctly point out) is to match entries that appear on the “Sold Properties” page with what is stored in the local DB. This would require a sophisticated regex in my opinion that would be able to match entries that may not be complete or entered incorrectly.

Do you think that this approach would work? More importantly would such a resource be useful to members of the Pin community?


#4

This is what salesporn.net does, except it uses LWP rather than Selenium so runs completely unattended. Results go into a database.

The problem with salesporn at the moment is that its sales state detection (for sale/sold/etc) is broken.


#5

Actually the Selenium Webdriver can run in headless mode as a cron job so it can be completely automated. Like the person who developed the salesporn.net site I think this can be done within a week by a competent developer. I myself am no ‘ninja developer’ either but would be willing to have a go at developing an initial basic version if people think that it would be of use to them as a tool.


#6

All data is useful! And the more people we have doing stuff with the data, the better. I say go for it!


#7

I use more of a sledgehammer approach. I just snapshot the whole of myhome.ie property for sale every three months. Starting last month my plan is to do it every month. I’m not that interested in asking price changes – there are sites that do that. My PPR match by address unfortunately has a hit rate of only about 15%. 30% of all addresses get discarded just because they have no house numbers. One of Mantissa’s very useful suggestions, apart from using Google geocoding which is much better at parsing addresses than my unfunded software is ever going to be, is to point out that you don’t need a pinpoint address – you just need a pair of addresses on myhome and the PPR to match better with each other than they do with any *other *address on the same databases.


#8

Agreed and that is why I think a good regex may be the key to getting the best possible match.


#9

I did some work matching PPR data with data from Daft and MyHome. The problem is matching the addresses is a bitch . I was able to match around 30% of the PPR properties I checked. The next version of the software will also include the size description of the property, from the property ad, and the sold price from the PPR , to get accurate price per square foot data. I matched around 1,000 properties . PM me if you are looking for data on specific areas . :slight_smile:
Here are some results :

10 Vale View Avenue, The Park, Cabinteely
46 Foxrock Avenue, Foxrock
Apt 56 Grand Central, Rockbrook, Sandyford Dublin 18
18 Glenbourne Crescent, Leopardstown Valley, Dublin 18
69 Gleann Na Ri, Druids Valley, Cabinteely
16 Glenbourne Grove, Leopardstown Valley


#10

I’ve got about 11,500 matches to the PPR. I’ve extracted the sizes, number of bedrooms, and property type from the myhome description. The description field has changed format a few times, but I’ve genericised the parsing algorithm and re-run it successfully on data back to 2011. Happy to share the regexes and approach ( – implementation is in Java).


#11

Very good, My aim of matching the data is to give buyers up-to-date price per Sq. foot data for sold properties in an area.
You would need to seperate the data by House, apartment etc. as this would skew the data horribly .
The Geocoding API looks very useful alright , as the PPR database is usually riddled with errors :slight_smile:
I am using Java as well , if you send me on what you are doing that would be great .The property ads usually stay up for a few months so if you run the program once a month that should cover the gamut alright.


#12

Here you go.

“BedsType” was the original css class name from the myhome site. Code’s a bit sloppy, hopefully clear what it does though. Just call the constructor, then any accessors you want. If getHouseType() returns null, it was a dud parse.


#13

Did anyone notice the sudden dip in national asking prices on Daftdrop for August??

daftdrop.com/#!graphView

I wonder is it a technical blip?


#14

I’ve so far experimentally geocoded 14,443 PPR addresses. This is in preparation for matching against historical myhome ad data.

13,414 (93%) resulted in a single geocode. However, many of these were townlands/administrative areas, so are still ambiguous.

7,250 single geocodes (50% of the overall total) included a ‘street_number’, i.e. a house number.

Another 2,117 single geocodes (15%) included a ‘premise’ (generally a house name) or ‘subpremise’ (apartment number).

That would imply 65% of the PPR addresses are in principle unambiguously matchable to corresponding myhome addresses. That’s slightly less than I got with my own homegrown address parsing, but I’d have more confidence in the Google geocodes. The actual match rate is likely to be vastly lower, on the order of 20%. I’m limited in the number of geocodes I can do per day, so this will be a long haul.


#15

Interesting.