City average: 41.1% off the peak
Top tier: 20.3% off the peak
Middle tier: 39.1% off the peak
Bottom tier: 47.9% off the peak
4% drop in one month in the top tier. Sales volume took a hard hit, too. The number of repeat sales in the top tier is down 50% month-to-month. This is probably the consequence of the stock market crash - that's where people with significant stock portfolios shop for their houses. Sales volumes in the other two tiers dropped by much smaller amounts - 10% and 14%.
Bottom tier as a whole is back to May '02, middle tier is back to September '02. 54-94 area solidified its position as the place with most severe price declines: a typical house in 54-94 is worth 48.4% of what it would've gone for at the peak in April-May '06.
Case-Shiller from 3 months into the future
Monday, December 15, 2008
Sunday, December 7, 2008
Neg-ams in San Diego
Downward pressure on house prices in San Diego has been in part caused by foreclosures. In the last few months, the supply of new foreclosures on the market has been reduced to a trickle. Mortgage modifications are getting increasingly popular. In a mortgage modification, lender agrees to lower the principal or the interest rate for a delinquent borrower and/or to transfer the modified loan to FHA. Jim the Realtor has a number of comments on this subject.
There's a catch. The lender will not reduce the principal much lower than the market value, and the borrower must qualify for a modified mortgage, fully documenting his income, with strict income limits.
And that's what brings us to neg-ams. A negative amortization loan is the one where monthly payment is insufficient to cover all interest accrued, making the principal grow rather than shrink over time. Neg-am borrowers' houses are way bigger and more expensive than they can realistically afford. Even after 30+% principal reductions, they are very unlikely to be able to pay modified mortgages. In addition, many neg-am borrowers were "investors" (owners of multiple non-owner-occupied properties), these obviously aren't getting any modifications. Areas with high densities of neg-ams will continue to produce foreclosures well into 2009 and possibly 2010. So, it's good to know where they are. Unfortunately, this kind of information is not readily available. We have to improvise.
I went through a database of all foreclosures that took place in San Diego County between July and November. There were approximately 10,000 of them. For every foreclosure, public sources give us 1) the initial loan amount, 2) the delinquency amount ( sum of all missed payments when the loan was declared in default ), 3) "sale amount" - the amount which the bank hopes to recover through foreclosure. #3 includes the balance on the loan on the day the borrower stopped paying, missed payments, late fees, and legal fees incurred by the bank. By subtracting #2 from #3, we can estimate the balance.
Inspection of the data quickly reveals a pattern. Half of all foreclosures have #3-#2 in the range between 101% and 104% of the initial balance. These are well known subprime interest-only loans. The bank needs to get back full amount loaned, and probably 1-4% of loan balance in legal fees.
Another big group of foreclosures has #3-#2 between 107% and 115% of the initial balance. These loans have lower interest rates. In this group, there's a noticeable correlation between the "sale amount" and the age of the loan. Loans in this category tend to grow 4-5% a year - the characteristic feature of a neg-am.
Overall, approximately a quarter of all foreclosures since July appear to have been caused by a neg-am. Neg-ams are everywhere. They are somewhat overrepresented in South Bay (44% of foreclosures in 91914, 37% in 91913, 32% in 92154). 92154 (San Ysidro) had the largest absolute number of suspected neg-ams in the dataset. 92037 (La Jolla) and 92109 (Pacific Beach) had relatively few foreclosures, but many of these were neg-ams. Weak prices in these areas should be expected to continue for some time, despite modification efforts.
Here's a random example of negative amortization. On 5/17/2006, Michael Ghanayem bought a 1,827 sf unit D102 in Seahaus condo complex in La Jolla, two blocks from the beach, for $1,090,000 with 10% down. Two months later, he refinanced into a $990,000 loan from "No Red Tape Mortgage" (later adding a $250k second from WaMu). On 11/17/2008, the condo was repossessed by the bank. Sale amount was $1,143,400. In two years, loan balance has grown at least 10%.
Here's another one. On 12/10/2003, Therese Ghanayem bought a house in a gated community in Santaluz (7378 Los Brazos) for $823,500. On 7/12/2005, the house was refinanced into two loans with Countrywide, $995k and $350k. First loan ended up in foreclosure on 8/15/2008. Sale amount was $1,133,294.
Both Seahaus D102 and 7378 Los Brazos are currently bank-owned and off the market. It's not clear what happened to Michael & Therese Ghanayem. With more than half a million tax-free dollars pulled out just from these two properties using cash-out refis, they can probably afford to retire in the Caribbean. (Incidentally, Michael is a 1998 Mt. Carmel High alumnus. That makes him 28 or so today.) But I digress.
Rural areas appear to have escaped this problem to some degree. Ramona (92065), north Escondido (92026), El Cajon (92020, 92021) have considerable numbers of foreclosures caused by "old-fashioned" subprime loans, but neg-am percentages in all these are below 20%.
There's a catch. The lender will not reduce the principal much lower than the market value, and the borrower must qualify for a modified mortgage, fully documenting his income, with strict income limits.
And that's what brings us to neg-ams. A negative amortization loan is the one where monthly payment is insufficient to cover all interest accrued, making the principal grow rather than shrink over time. Neg-am borrowers' houses are way bigger and more expensive than they can realistically afford. Even after 30+% principal reductions, they are very unlikely to be able to pay modified mortgages. In addition, many neg-am borrowers were "investors" (owners of multiple non-owner-occupied properties), these obviously aren't getting any modifications. Areas with high densities of neg-ams will continue to produce foreclosures well into 2009 and possibly 2010. So, it's good to know where they are. Unfortunately, this kind of information is not readily available. We have to improvise.
I went through a database of all foreclosures that took place in San Diego County between July and November. There were approximately 10,000 of them. For every foreclosure, public sources give us 1) the initial loan amount, 2) the delinquency amount ( sum of all missed payments when the loan was declared in default ), 3) "sale amount" - the amount which the bank hopes to recover through foreclosure. #3 includes the balance on the loan on the day the borrower stopped paying, missed payments, late fees, and legal fees incurred by the bank. By subtracting #2 from #3, we can estimate the balance.
Inspection of the data quickly reveals a pattern. Half of all foreclosures have #3-#2 in the range between 101% and 104% of the initial balance. These are well known subprime interest-only loans. The bank needs to get back full amount loaned, and probably 1-4% of loan balance in legal fees.
Another big group of foreclosures has #3-#2 between 107% and 115% of the initial balance. These loans have lower interest rates. In this group, there's a noticeable correlation between the "sale amount" and the age of the loan. Loans in this category tend to grow 4-5% a year - the characteristic feature of a neg-am.
Overall, approximately a quarter of all foreclosures since July appear to have been caused by a neg-am. Neg-ams are everywhere. They are somewhat overrepresented in South Bay (44% of foreclosures in 91914, 37% in 91913, 32% in 92154). 92154 (San Ysidro) had the largest absolute number of suspected neg-ams in the dataset. 92037 (La Jolla) and 92109 (Pacific Beach) had relatively few foreclosures, but many of these were neg-ams. Weak prices in these areas should be expected to continue for some time, despite modification efforts.
Here's a random example of negative amortization. On 5/17/2006, Michael Ghanayem bought a 1,827 sf unit D102 in Seahaus condo complex in La Jolla, two blocks from the beach, for $1,090,000 with 10% down. Two months later, he refinanced into a $990,000 loan from "No Red Tape Mortgage" (later adding a $250k second from WaMu). On 11/17/2008, the condo was repossessed by the bank. Sale amount was $1,143,400. In two years, loan balance has grown at least 10%.
Here's another one. On 12/10/2003, Therese Ghanayem bought a house in a gated community in Santaluz (7378 Los Brazos) for $823,500. On 7/12/2005, the house was refinanced into two loans with Countrywide, $995k and $350k. First loan ended up in foreclosure on 8/15/2008. Sale amount was $1,133,294.
Both Seahaus D102 and 7378 Los Brazos are currently bank-owned and off the market. It's not clear what happened to Michael & Therese Ghanayem. With more than half a million tax-free dollars pulled out just from these two properties using cash-out refis, they can probably afford to retire in the Caribbean. (Incidentally, Michael is a 1998 Mt. Carmel High alumnus. That makes him 28 or so today.) But I digress.
Rural areas appear to have escaped this problem to some degree. Ramona (92065), north Escondido (92026), El Cajon (92020, 92021) have considerable numbers of foreclosures caused by "old-fashioned" subprime loans, but neg-am percentages in all these are below 20%.
Saturday, November 22, 2008
Election map
Perhaps you'll find this interesting:
Election results are from San Diego Registrar of Voters; precinct boundaries are from SanGIS, parsed with Shapefile library. I used the map of Congressional Districts as the background - that was the highest-resolution map of San Diego I could find. (I tried to stitch together a map from Google Maps screenshots, but that didn't work so well - I couldn't get the election overlay to match geographic features perfectly.)
All precincts with 20 or more recorded votes are shown.
It's striking how little opposition there is for Prop 8 outside North County coast and central neighborhoods (Hillcrest, South Park & such). Most neighborhoods south of 94 vote for Obama and Yes on 8. Same with Indian reservations (Pala, Barona, Santa Ysabel) and democratic pockets in Escondido and El Cajon.
Election results are from San Diego Registrar of Voters; precinct boundaries are from SanGIS, parsed with Shapefile library. I used the map of Congressional Districts as the background - that was the highest-resolution map of San Diego I could find. (I tried to stitch together a map from Google Maps screenshots, but that didn't work so well - I couldn't get the election overlay to match geographic features perfectly.)
All precincts with 20 or more recorded votes are shown.
It's striking how little opposition there is for Prop 8 outside North County coast and central neighborhoods (Hillcrest, South Park & such). Most neighborhoods south of 94 vote for Obama and Yes on 8. Same with Indian reservations (Pala, Barona, Santa Ysabel) and democratic pockets in Escondido and El Cajon.
Sunday, November 9, 2008
Mandatory election post
Every blogger seems to have something to say about our recent election. Here's my try.
This is how different parts of San Diego County voted on November 4th.
All numbers come from here.
Rural areas are pro-McCain, central San Diego neighborhoods (e.g. Golden Hill, South Park) are heavily pro-Obama. Hillcrest and North Park favor Obama 5 to 1. In Fairbanks Ranch and Rancho Santa Fe, 67% of votes went for McCain, but neighboring Del Mar Heights and Solana Beach vote 57-59% for Obama. (Carmel Valley is probably part of Del Mar Heights in their definition)
Hillcrest is the most liberal, deep East County (Campo, Descanso, Boulevard) is the most conservative. Chollas Park, Encanto, San Ysidro, National City vote for Obama and Yes on 8 at the same time.
Next, I'll try to build a map. I'll either have to find 2008 precinct definitions or adapt 2006 definitions somehow.
This is how different parts of San Diego County voted on November 4th.
Neighborhood | Votes for McCain | Votes for Obama | % for Obama | % Yes on Prop 8 |
Alpine | 4202 | 1939 | 31.6% | 69.4% |
Ballena | 310 | 133 | 30.0% | 69.4% |
Bonita | 2748 | 2427 | 46.9% | 62.6% |
Bonsall | 690 | 358 | 34.2% | 65.9% |
Borrego Springs | 477 | 518 | 52.1% | 56.3% |
Bostonia | 3786 | 2755 | 42.1% | 66.3% |
Boulevard | 341 | 164 | 32.5% | 72.8% |
Buena | 1845 | 1390 | 43.0% | 62.0% |
Campo | 414 | 159 | 27.7% | 77.4% |
Carlsbad | 21637 | 21843 | 50.2% | 51.2% |
Centre City | 1893 | 4165 | 68.8% | 36.5% |
Chollas Park | 716 | 4571 | 86.5% | 65.0% |
Chula Vista | 24675 | 36318 | 59.5% | 62.5% |
City Heights | 1059 | 3284 | 75.6% | 49.9% |
Clairemont E | 6240 | 7731 | 55.3% | 51.8% |
Clairemont N | 4458 | 5993 | 57.3% | 47.7% |
Clairemont S | 3605 | 5363 | 59.8% | 43.3% |
Coronado | 4224 | 3095 | 42.3% | 54.8% |
Crest | 840 | 437 | 34.2% | 69.2% |
Dehesa | 195 | 132 | 40.4% | 63.5% |
Del Dios | 2372 | 2151 | 47.6% | 58.1% |
Del Mar | 812 | 1439 | 63.9% | 32.6% |
Del Mar Heights | 5604 | 8362 | 59.9% | 40.3% |
Descanso | 1134 | 474 | 29.5% | 70.1% |
El Cajon | 13004 | 11324 | 46.5% | 63.9% |
Elfin Forest | 124 | 104 | 45.6% | 55.3% |
Encanto E | 2857 | 5790 | 67.0% | 68.7% |
Encanto W | 1660 | 6356 | 79.3% | 66.5% |
Encinitas | 9981 | 16538 | 62.4% | 38.3% |
Escondido | 18635 | 15711 | 45.7% | 64.1% |
Fairbanks Ranch | 503 | 250 | 33.2% | 58.7% |
Fallbrook | 8732 | 4816 | 35.5% | 68.1% |
Felicita Park | 892 | 499 | 35.9% | 62.0% |
Foothill | 287 | 219 | 43.3% | 60.6% |
Golden Hill | 413 | 2098 | 83.6% | 30.7% |
Harbison Canyon | 301 | 147 | 32.8% | 63.2% |
Harbor | 745 | 643 | 46.3% | 50.1% |
Helix | 3030 | 2752 | 47.6% | 55.8% |
Hidden Meadows | 2131 | 1076 | 33.6% | 69.1% |
Hillcrest | 1481 | 7791 | 84.0% | 16.4% |
Imperial Beach | 2553 | 3511 | 57.9% | 55.9% |
Jacumba | 58 | 73 | 55.7% | 65.6% |
Jamacha | 661 | 373 | 36.1% | 67.0% |
Jamul | 1963 | 1139 | 36.7% | 66.7% |
Julian | 789 | 504 | 39.0% | 60.1% |
La Jolla | 6467 | 8886 | 57.9% | 40.0% |
La Mesa | 9133 | 11730 | 56.2% | 50.7% |
La Playa | 1929 | 2004 | 51.0% | 47.9% |
Lake Morena | 331 | 153 | 31.6% | 71.4% |
Lake San Marcos | 1515 | 927 | 38.0% | 65.9% |
Lakeside | 7590 | 3868 | 33.8% | 69.0% |
Lakeview Flinn | 3491 | 1634 | 31.9% | 70.4% |
Lemon Grove | 2919 | 4114 | 58.5% | 61.4% |
Lincoln Acres | 172 | 319 | 65.0% | 65.7% |
Linda Vista | 2787 | 4488 | 61.7% | 48.9% |
Loma Portal | 2668 | 3408 | 56.1% | 44.9% |
Lomas Santa Fe | 231 | 314 | 57.6% | 36.8% |
Middletown | 1863 | 5030 | 73.0% | 28.1% |
Midway Old Town | 1498 | 2734 | 64.6% | 40.0% |
Mira Mesa | 9464 | 11220 | 54.2% | 57.2% |
Mission Bay | 659 | 1087 | 62.3% | 33.8% |
Mission Hills | 1286 | 3220 | 71.5% | 27.7% |
Mission Valley | 1459 | 2966 | 67.0% | 35.1% |
Montezuma | 2004 | 4526 | 69.3% | 33.8% |
Montgomery | 296 | 449 | 60.3% | 44.6% |
National City | 3273 | 6065 | 64.9% | 65.9% |
Navajo | 6324 | 7238 | 53.4% | 50.5% |
Nestor | 3488 | 6460 | 64.9% | 65.0% |
Normal Heights | 3227 | 10063 | 75.7% | 34.4% |
North Park | 2322 | 11834 | 83.6% | 23.5% |
Oak Grove | 268 | 188 | 41.2% | 68.1% |
Ocean Beach | 2217 | 6328 | 74.1% | 26.9% |
Oceanside | 25385 | 26334 | 50.9% | 59.5% |
Otay | 2170 | 3655 | 62.7% | 66.4% |
Pacific Beach | 5390 | 11021 | 67.2% | 31.6% |
Pala | 57 | 115 | 66.9% | 59.7% |
Paradise Hills | 3374 | 4808 | 58.8% | 68.4% |
Pauma Valley | 174 | 205 | 54.1% | 62.9% |
Pomerado | 8297 | 8830 | 51.6% | 52.1% |
Potrero | 161 | 140 | 46.5% | 75.1% |
Poway | 11416 | 7920 | 41.0% | 60.0% |
Rainbow | 287 | 200 | 41.1% | 65.7% |
Ramona | 8215 | 3886 | 32.1% | 68.0% |
Ranchita | 156 | 109 | 41.1% | 64.4% |
Rancho Bernardo | 9423 | 8625 | 47.8% | 57.4% |
Rancho El Cajon | 1453 | 637 | 30.5% | 72.3% |
Rancho Santa Fe | 1652 | 837 | 33.6% | 57.7% |
Rho Monserate | 432 | 247 | 36.4% | 72.1% |
Rho Penasquitos | 14972 | 16475 | 52.4% | 52.7% |
Rincon D Diablo | 2045 | 1188 | 36.7% | 64.4% |
Rock Springs | 431 | 267 | 38.3% | 65.3% |
Rolando Redwood | 3380 | 7575 | 69.1% | 51.4% |
San Carlos | 5361 | 5863 | 52.2% | 51.7% |
San Marcos | 11446 | 11181 | 49.4% | 59.8% |
San Ysidro | 1709 | 4217 | 71.2% | 65.4% |
Santee | 12007 | 7869 | 39.6% | 62.7% |
Se San Diego E | 339 | 2116 | 86.2% | 65.4% |
Se San Diego W | 289 | 1838 | 86.4% | 61.0% |
Serra Mesa | 3130 | 3845 | 55.1% | 52.0% |
Solana Beach | 2571 | 3445 | 57.3% | 41.1% |
South Park | 1400 | 5628 | 80.1% | 25.2% |
Spring Valley | 7093 | 8756 | 55.2% | 64.1% |
Sweetwater | 405 | 415 | 50.6% | 66.6% |
Tierrasanta | 4455 | 5437 | 55.0% | 48.2% |
University N | 3812 | 11904 | 75.7% | 29.7% |
University S | 2791 | 4063 | 59.3% | 43.8% |
Valle De Oro | 6877 | 4506 | 39.6% | 63.6% |
Valley Center | 3869 | 1806 | 31.8% | 70.2% |
Vista | 11208 | 10934 | 49.4% | 60.8% |
Vista Acres | 724 | 556 | 43.4% | 61.6% |
Whispering Palms | 804 | 484 | 37.6% | 56.1% |
Yaqui Wells | 61 | 64 | 51.2% | 57.3% |
All numbers come from here.
Rural areas are pro-McCain, central San Diego neighborhoods (e.g. Golden Hill, South Park) are heavily pro-Obama. Hillcrest and North Park favor Obama 5 to 1. In Fairbanks Ranch and Rancho Santa Fe, 67% of votes went for McCain, but neighboring Del Mar Heights and Solana Beach vote 57-59% for Obama. (Carmel Valley is probably part of Del Mar Heights in their definition)
Hillcrest is the most liberal, deep East County (Campo, Descanso, Boulevard) is the most conservative. Chollas Park, Encanto, San Ysidro, National City vote for Obama and Yes on 8 at the same time.
Next, I'll try to build a map. I'll either have to find 2008 precinct definitions or adapt 2006 definitions somehow.
Friday, November 7, 2008
San Diego HPI - October '08
City average: 37.9% off the peak
Top tier: 15.5% off the peak
Middle tier: 37.4% off the peak
Bottom tier: 45.9% off the peak
These numbers represent transactions that went into escrow in late August or early September. Stock market began tanking in the last week of September. I expect to see some impact of the stock market crash in November and December data.
Top tier: 15.5% off the peak
Middle tier: 37.4% off the peak
Bottom tier: 45.9% off the peak
These numbers represent transactions that went into escrow in late August or early September. Stock market began tanking in the last week of September. I expect to see some impact of the stock market crash in November and December data.
Wednesday, October 1, 2008
San Diego HPI - September '08
City average: 37.5% off the peak
Top tier: 15.3% off the peak
Middle tier: 37.6% off the peak
Bottom tier: 44.9% off the peak
Note: the chart is rebased to use December '99 = 100%, this way my numbers can be directly compared with Case-Shiller.
Top tier: 15.3% off the peak
Middle tier: 37.6% off the peak
Bottom tier: 44.9% off the peak
Note: the chart is rebased to use December '99 = 100%, this way my numbers can be directly compared with Case-Shiller.
Month | S&P Case-Shiller | SDHPI |
January '08 | 197.45 | 198.06 |
February '08 | 190.34 | 190.64 |
March '08 | 185.42 | 184.87 |
April '08 | 180.56 | 179.99 |
May '08 | 178.03 | 176.52 |
June '08 | 175.37 | 174.37 |
July '08 | 172.20 | 171.42 |
August '08 | 168.25 | |
September '08 | 164.32 | |
October '08 | 159.23 |
Monday, September 15, 2008
Good schools vs bad schools
This is only marginally related to the main topic of my blog.
When I do tiered price analysis, I define "top tier" as a set of desirable geographical areas with low crime and good schools. But what makes good schools good?
This is a plot of API scores of 450 elementary schools in San Diego County against average education level of all parents of students in these schools. Average parent education level is defined in such a way that it would be 5 if every student in a school had at least one parent who spent any time in a graduate school; it would be 1 if no one in the school had a single parent who managed to graduate from high school.
It seems that parent education is a strong predictor of school ratings. On the other hand, school district itself has almost no say.
All numbers come from here
http://www.cde.ca.gov/ta/ac/ap/apidatafiles.asp
When I do tiered price analysis, I define "top tier" as a set of desirable geographical areas with low crime and good schools. But what makes good schools good?
This is a plot of API scores of 450 elementary schools in San Diego County against average education level of all parents of students in these schools. Average parent education level is defined in such a way that it would be 5 if every student in a school had at least one parent who spent any time in a graduate school; it would be 1 if no one in the school had a single parent who managed to graduate from high school.
It seems that parent education is a strong predictor of school ratings. On the other hand, school district itself has almost no say.
All numbers come from here
http://www.cde.ca.gov/ta/ac/ap/apidatafiles.asp
Monday, September 1, 2008
San Diego HPI - August '08
City average: 35.1% off the peak
Top tier: 14.3% off the peak
Middle tier: 35.3% off the peak
Bottom tier: 44.1% off the peak
Almost all regions are showing declines again. City average is down 2% month-to-month, top tier is at the lowest level since 2004, and 54-94 geo finally went below 50% off the peak.
Price changes since 2000:
Top tier: 14.3% off the peak
Middle tier: 35.3% off the peak
Bottom tier: 44.1% off the peak
Almost all regions are showing declines again. City average is down 2% month-to-month, top tier is at the lowest level since 2004, and 54-94 geo finally went below 50% off the peak.
Price changes since 2000:
Friday, August 22, 2008
Forecast
This is an attempt to predict the dynamics of the real estate market in San Diego over the next 3-4 years.
"Fundamentals" curve assumes more or less unchanged interest rates and steady wage inflation that averages 3%/year going forward. For the top tier I use a soft-landing scenario (exponential decay to fundamentals). To extrapolate two other tiers, I use quadratic approximations of season-adjusted HPI points from the last 9 months.
Summary of predictions:
* Aggregate Case-Shiller bottom in February '09 in 150-155 range (late '02 pricing)
* Top tier: 5% decline followed by many years of scraping along the bottom
* Middle tier: 6-8% decline followed by a mild bounce-back
Caveats:
* Bottom tier may get some support from FHA short-refinancing program, in which case the decline will not be so severe.
* I'm assuming that GSE conforming-jumbo loans and higher FHA loan limits stay with us until the return of a healthy jumbo market. Technically, they are supposed to be discontinued on Dec 31, 2008, but it's likely that loan limits will be extended.
* Interest rates may not stay the same. Here are two alternative scenarios, with rates heading to 5.5% and 7.5% long term, respectively:
I'm not a prophet and I may be seriously off, only time will tell.
UPDATE: recalculated the "fundamentals" curve using actual CPI and mortgage rate data.
"Fundamentals" curve assumes more or less unchanged interest rates and steady wage inflation that averages 3%/year going forward. For the top tier I use a soft-landing scenario (exponential decay to fundamentals). To extrapolate two other tiers, I use quadratic approximations of season-adjusted HPI points from the last 9 months.
Summary of predictions:
* Aggregate Case-Shiller bottom in February '09 in 150-155 range (late '02 pricing)
* Top tier: 5% decline followed by many years of scraping along the bottom
* Middle tier: 6-8% decline followed by a mild bounce-back
Caveats:
* Bottom tier may get some support from FHA short-refinancing program, in which case the decline will not be so severe.
* I'm assuming that GSE conforming-jumbo loans and higher FHA loan limits stay with us until the return of a healthy jumbo market. Technically, they are supposed to be discontinued on Dec 31, 2008, but it's likely that loan limits will be extended.
* Interest rates may not stay the same. Here are two alternative scenarios, with rates heading to 5.5% and 7.5% long term, respectively:
I'm not a prophet and I may be seriously off, only time will tell.
UPDATE: recalculated the "fundamentals" curve using actual CPI and mortgage rate data.
Tuesday, August 5, 2008
Prices by zip code: 2008 vs 2002
Approximate median values of detached homes in each community in Q2 2008, and how much they were worth in Q3 2002.
Rank | Area | Definition | Median price | Median price - Q3/02 | Rank - Q3/02 | Change |
1 | Logan Heights | 92113 | $186,000 | $195,000 | 1 | -5% |
2 | Chollas/Euclid | 92105 | $242,000 | $233,000 | 2 | 4% |
3 | Encanto | 92114 | $250,000 | $245,000 | 4 | 2% |
4 | Golden Hill | 92102 | $253,000 | $263,000 | 5 | -4% |
5 | National City | 91950 | $263,000 | $234,000 | 3 | 12% |
6 | Vista NW | 92083 | $267,000 | $267,000 | 7 | 0% |
7 | Lemon Grove | 91945 | $281,000 | $273,000 | 8 | 3% |
8 | Escondido E | 92027 | $289,000 | $282,000 | 10 | 2% |
9 | Spring Valley | 91977,91978 | $290,000 | $278,000 | 9 | 4% |
10 | Escondido SE (valley) | 92025 | $290,000 | $283,000 | 11 | 2% |
11 | Paradise Hills | 92139 | $290,000 | $267,000 | 6 | 9% |
12 | Oceanside NE | 92057 | $318,000 | $289,000 | 12 | 10% |
13 | Santee | 92071 | $329,000 | $297,000 | 14 | 11% |
14 | San Ysidro | 92173 | $331,000 | $303,000 | 18 | 9% |
15 | Escondido N | 92026 | $332,000 | $306,000 | 20 | 8% |
16 | Chula Vista SW | 91911 | $334,000 | $298,000 | 16 | 12% |
17 | El Cajon E | 92021 | $335,000 | $299,000 | 17 | 12% |
18 | NE Vista | 92084 | $337,000 | $327,000 | 27 | 3% |
19 | Otay Mesa | 92154 | $347,000 | $298,000 | 15 | 16% |
20 | El Cajon W | 92020 | $348,000 | $295,000 | 13 | 18% |
21 | Oceanside SE | 92056 | $354,000 | $304,000 | 19 | 16% |
22 | Imperial Beach | 91932 | $360,000 | $319,000 | 23 | 13% |
23 | La Mesa | 91941,91942 | $360,000 | $318,000 | 22 | 13% |
24 | Chula Vista NW | 91910 | $374,000 | $340,000 | 31 | 10% |
25 | Lakeside | 92040 | $375,000 | $319,000 | 24 | 18% |
26 | Linda Vista | 92111 | $378,000 | $345,000 | 32 | 10% |
27 | San Marcos N | 92069 | $382,000 | $337,000 | 30 | 13% |
28 | SDSU | 92115 | $385,000 | $320,000 | 25 | 20% |
29 | Rancho San Diego | 92019 | $386,000 | $354,000 | 36 | 9% |
30 | Mira Mesa | 92126 | $389,000 | $328,000 | 28 | 19% |
31 | W Oceanside | 92054 | $392,000 | $317,000 | 21 | 24% |
32 | San Carlos | 92119 | $394,000 | $346,000 | 33 | 14% |
33 | Serra Mesa | 92123 | $399,000 | $324,000 | 26 | 23% |
34 | Ramona | 92065 | $402,000 | $350,000 | 35 | 15% |
35 | Eastlake, Otay Ranch | 91913,91915 | $416,000 | $356,000 | 37 | 17% |
36 | Vista S | 92081 | $417,000 | $328,000 | 29 | 27% |
37 | Falbrook | 92028 | $434,000 | $350,000 | 34 | 24% |
38 | Clairemont | 92117 | $441,000 | $374,000 | 40 | 18% |
39 | Alpine | 91901 | $454,000 | $410,000 | 47 | 11% |
40 | San Marcos S (except San Elijo Hills) | 92078 | $455,000 | $369,000 | 39 | 23% |
41 | Valley Center | 92082 | $457,000 | $407,000 | 45 | 12% |
42 | Normal Heights | 92116 | $457,000 | $379,000 | 42 | 21% |
43 | Eastlake NE | 91914 | $482,000 | $409,000 | 46 | 18% |
44 | Allied Gardens, Del Cerro | 92120 | $485,000 | $375,000 | 41 | 29% |
45 | South Park | 92104 | $491,000 | $365,000 | 38 | 35% |
46 | SW Escondido | 92029 | $495,000 | $417,000 | 48 | 19% |
47 | Tierrasanta | 92124 | $513,000 | $420,000 | 51 | 22% |
48 | Poway | 92064 | $514,000 | $404,000 | 44 | 27% |
49 | Rancho Bernardo | 92127 | $528,000 | $394,000 | 43 | 34% |
50 | Bonita | 91902 | $536,000 | $434,000 | 53 | 24% |
51 | Carlsbad NE | 92010 | $547,000 | $420,000 | 50 | 30% |
52 | Penasquitos | 92129 | $560,000 | $418,000 | 49 | 34% |
53 | Carmel Mtn Ranch | 92128 | $571,000 | $433,000 | 52 | 32% |
54 | San Elijo Hills | 92078 | $592,000 | $452,000 | 54 | 31% |
55 | Bay Park | 92110 | $593,000 | $464,000 | 55 | 28% |
56 | Escondido SE (hills) | 92025 | $644,000 | $504,000 | 59 | 28% |
57 | Carlsbad NW | 92008 | $666,000 | $485,000 | 57 | 37% |
58 | Scripps Ranch | 92131 | $676,000 | $520,000 | 60 | 30% |
59 | University City | 92122 | $677,000 | $474,000 | 56 | 43% |
60 | Ocean Beach | 92107 | $681,000 | $496,000 | 58 | 37% |
61 | Hillcrest, Mission Hills | 92103 | $735,000 | $559,000 | 64 | 31% |
62 | Carlsbad SE | 92009 | $743,000 | $527,000 | 61 | 41% |
63 | Carlsbad SW | 92011 | $760,000 | $551,000 | 63 | 38% |
64 | Pacific Beach | 92109 | $795,000 | $563,000 | 65 | 41% |
65 | Encinitas | 92024 | $804,000 | $576,000 | 66 | 40% |
66 | 4S Ranch | 92127 | $838,000 | $582,000 | 67 | 44% |
67 | Point Loma | 92106 | $875,000 | $616,000 | 68 | 42% |
68 | Cardiff | 92007 | $884,000 | $538,000 | 62 | 64% |
69 | Carmel Valley | 92130 | $982,000 | $690,000 | 69 | 42% |
70 | Del Mar, Solana Beach | 92014,92075 | $1,393,000 | $912,000 | 70 | 53% |
71 | La Jolla | 92037 | $1,698,000 | $1,119,000 | 72 | 52% |
72 | Coronado | 92118 | $1,748,000 | $1,108,000 | 71 | 58% |
73 | Rancho Santa Fe | 92067,92091 | $2,506,000 | $1,932,000 | 73 | 30% |
Friday, August 1, 2008
San Diego HPI - July '08
City average: 33.8% off the peak
Top tier: 13.3% off the peak
Middle tier: 34.2% off the peak
Bottom tier: 42.6% off the peak
Steady decline continues. Top tier remains resilient, but the spring bounce is clearly over. 54-94 geo is getting close to breaking the 50% off the peak.
Two lowest-priced geos ("54-94" and "north of 78") are back to mid-'02 prices. If this trend persists, we'll see 2001 prices before the end of the year.
These charts include all June closing late-reporters. Apparent stabilization in several neighborhoods in June charts was a statistical fluke.
Top tier: 13.3% off the peak
Middle tier: 34.2% off the peak
Bottom tier: 42.6% off the peak
Steady decline continues. Top tier remains resilient, but the spring bounce is clearly over. 54-94 geo is getting close to breaking the 50% off the peak.
Two lowest-priced geos ("54-94" and "north of 78") are back to mid-'02 prices. If this trend persists, we'll see 2001 prices before the end of the year.
These charts include all June closing late-reporters. Apparent stabilization in several neighborhoods in June charts was a statistical fluke.
Tuesday, July 1, 2008
San Diego HPI - June '08
City average: 32.0% off the peak
Top tier: 11.8% off the peak
Middle tier: 33.0% off the peak
Bottom tier: 40.4% off the peak
Low and middle tiers are quite reasonably priced. If this were January, I'd probably call the bottom. But this is June, and we have record-shattering numbers of defaults in the pipeline, and these defaults will mature into REOs during fall and winter, when real estate activity is traditionally low. Also, interest rates have gone up considerably in the last month. Higher interest rates wouldn't yet be reflected in June closings. I think there's still potential on the downside.
Interest rates are about as high today as they were in August-September of '02 (except jumbos). If we assume that houses were fairly valued back then, and add 3% average annual inflation, we get this for current valuations:
Carmel Valley, 4S, Scripps Ranch: 40% overpriced
Carlsbad, Encinitas: 36% overpriced
Rancho Bernardo, Rancho Penasquitos: 28% overpriced (except for lower-end pockets where you can get by without a jumbo)
Mission Trails (Del Cerro, Allied Gardens, Tierrasanta): 6% to 22% overpriced
Clairemont, Mira Mesa: 3% underpriced
Eastlake, Otay Ranch: 8% underpriced (except for high-end parts of 91914)
Chula Vista, Imperial Beach: 9% underpriced
North of 78 (east Oceanside, north Vista): 10% underpriced
54-94: 15% underpriced
Top tier: 11.8% off the peak
Middle tier: 33.0% off the peak
Bottom tier: 40.4% off the peak
Low and middle tiers are quite reasonably priced. If this were January, I'd probably call the bottom. But this is June, and we have record-shattering numbers of defaults in the pipeline, and these defaults will mature into REOs during fall and winter, when real estate activity is traditionally low. Also, interest rates have gone up considerably in the last month. Higher interest rates wouldn't yet be reflected in June closings. I think there's still potential on the downside.
Interest rates are about as high today as they were in August-September of '02 (except jumbos). If we assume that houses were fairly valued back then, and add 3% average annual inflation, we get this for current valuations:
Carmel Valley, 4S, Scripps Ranch: 40% overpriced
Carlsbad, Encinitas: 36% overpriced
Rancho Bernardo, Rancho Penasquitos: 28% overpriced (except for lower-end pockets where you can get by without a jumbo)
Mission Trails (Del Cerro, Allied Gardens, Tierrasanta): 6% to 22% overpriced
Clairemont, Mira Mesa: 3% underpriced
Eastlake, Otay Ranch: 8% underpriced (except for high-end parts of 91914)
Chula Vista, Imperial Beach: 9% underpriced
North of 78 (east Oceanside, north Vista): 10% underpriced
54-94: 15% underpriced
Monday, June 16, 2008
Bay Area
"880": Pleasanton, Dublin, San Ramon, Walnut Creek, Pleasant Hill
"SE": Fremont, Milpitas, northernmost San Jose
"Ghetto" zip codes (e.g. Concord, Union City) are excluded.
Trulia's resale database is very limited, so 3 month resolution with a lot of noise is the best I can do. I planned to add a series for SW (Cupertino, Palo Alto & such) but there was not enough data, and the resulting chart was too erratic. It does appear that southwest Bay Area is still stuck at 2005 pricing.
Tuesday, June 10, 2008
Three suburbs
I looked and looked at today's post on BMIT and in the end I decided that it could use some hard HPI data.
And here it is.
I basically took HPI profiles for three suburbs and scaled them using known sale prices of those three houses.
Some thoughts. (Read the original post to understand what's going on)
The Temecula house is slightly above median for Temecula-Murrieta area, and it's the nicest of all three. Notice the 10k sf lot and the fact that it has no HOA or Mello-Roos. It is aggressively priced and should not have much trouble selling.
The 4S house is well below median for 4S. It's an 'economy' house. In 4S Ranch and Carmel Valley, everyone and their mother has 3000 sq.ft., and that's not the factor that determines value of the house. The key factor is lot size. BMIT post incorrectly says that 17004 Ralphs Ranch Rd. has 8276 sq.ft. lot. 'Economy' 4S houses never have lots that big. It's really 5300 sf. There may be something wrong with interior and upgrades. Here's a good example of a 'median' 4S house:
15161 Cross Stone Dr, San Diego CA 92127
4 beds, 2.5 baths, 3,031 sq ft, 6,720 sf. Sold for $812,500 on 03/20/2008. 17004 Ralphs Ranch was already listed, and the buyer of 15161 Cross Stone surely must have looked at it before committing to pay 130K more for a different house.
The CV house is below median (it's also on a postage stamp lot). At 920k, it's somewhat overpriced. In today's market, it would have a better chance of selling at 860k.
If CV and 4S were deflated to the same level as Temecula, 17004 Ralphs Ranch would be worth 500k, and 13854 Kerry would be worth 650k.
And here it is.
I basically took HPI profiles for three suburbs and scaled them using known sale prices of those three houses.
Some thoughts. (Read the original post to understand what's going on)
The Temecula house is slightly above median for Temecula-Murrieta area, and it's the nicest of all three. Notice the 10k sf lot and the fact that it has no HOA or Mello-Roos. It is aggressively priced and should not have much trouble selling.
The 4S house is well below median for 4S. It's an 'economy' house. In 4S Ranch and Carmel Valley, everyone and their mother has 3000 sq.ft., and that's not the factor that determines value of the house. The key factor is lot size. BMIT post incorrectly says that 17004 Ralphs Ranch Rd. has 8276 sq.ft. lot. 'Economy' 4S houses never have lots that big. It's really 5300 sf. There may be something wrong with interior and upgrades. Here's a good example of a 'median' 4S house:
15161 Cross Stone Dr, San Diego CA 92127
4 beds, 2.5 baths, 3,031 sq ft, 6,720 sf. Sold for $812,500 on 03/20/2008. 17004 Ralphs Ranch was already listed, and the buyer of 15161 Cross Stone surely must have looked at it before committing to pay 130K more for a different house.
The CV house is below median (it's also on a postage stamp lot). At 920k, it's somewhat overpriced. In today's market, it would have a better chance of selling at 860k.
If CV and 4S were deflated to the same level as Temecula, 17004 Ralphs Ranch would be worth 500k, and 13854 Kerry would be worth 650k.
Sunday, June 8, 2008
Temecula & Murrieta
This was on my back burner for a long time. There's no equivalent of SDLookup or San Diego County Real Estate Information in Riverside County (as far as I know), so it's harder to get the data. In the end I went with Trulia - they have about 9 months' worth of complete sales records and historical data for everything sold during this 9 month window.
As of May '08, Temecula and Murrieta are 40% off the peak, with no indications of slowing down.
As of May '08, Temecula and Murrieta are 40% off the peak, with no indications of slowing down.
Friday, June 6, 2008
SDHPI vs. Case-Shiller
Case-Shiller is shifted back one month: the most recent point ("March 2008", published May 25) is plotted in February on the chart.
The agreement seems to be quite good, SDHPI is a bit more "bearish" than C-S, possibly because C-S assigns higher weights to expensive houses and SDHPI weighs everyone equally.
Even better agreement in low and middle tiers, clear discrepancy in the high tier. It's mainly because SDHPI's high tier is geographical and C-S's high tier is price based. Larger properties from middle and even low SDHPI tiers occasionally end up in C-S's high tier. Clearly, price stability has more to do with geographic factors (schools, crime, demographics) than with absolute price values.
Monday, June 2, 2008
May
City average: 31.2% off the peak, July 2003
Top tier: 11.9% off the peak, April 2004
Middle tier: 32.5% off the peak, June 2003
Bottom tier: 39.6% off the peak, February 2003
Slight change of zone definitions: "north of 78" is defined as 92056,92057,92083,92084. San Marcos North (92069) fares slightly better and 92078 fares better yet. Escondido is very heterogeneous and "railroad tracks" do not match with zip code boundaries.
Top tier: 11.9% off the peak, April 2004
Middle tier: 32.5% off the peak, June 2003
Bottom tier: 39.6% off the peak, February 2003
Slight change of zone definitions: "north of 78" is defined as 92056,92057,92083,92084. San Marcos North (92069) fares slightly better and 92078 fares better yet. Escondido is very heterogeneous and "railroad tracks" do not match with zip code boundaries.
Monday, May 19, 2008
Is this the bottom?
San Diego Union Tribune ran an article claiming that San Diego county housing is bottoming. A couple of weeks ago WSJ made a similar claim: "...it is very likely that April 2008 will mark the bottom of the U.S. housing market."
Is it, or is it not?
* April is not a likely month to mark the bottom. Barring major interest rate moves or bailouts (which we did not have), typical seasonal patterns suggest that the bottom (as observed by a HPI) should occur in January or February. If you use Case-Shiller, you'll see it in February or March because its results are delayed by one month. Seasonal bounce can temporarily override the correction. It did that every year during the 90's bust. But, if it does not manage to do that till April, the correction is probably not over.
90's San Diego "false bottoms":
1993: May
1994: February
1995: February
True bottom: February of 1996
* Housing price uptick is only seen in median prices, not in this HPI. Median prices are notorious for their high noise levels. Whether we see an uptick in May, remains to be seen.
* Low and middle tier are close to fundamentals, but we're going through never-before-seen numbers of foreclosures (dwarfing anything recorded during the 90's bust) and those are likely to maintain some downward pressure on prices.
* High tier is nowhere close to fundamentals and '04-'05 vintage neg-ams are yet to reset to fully amortizing.
Based on current dynamics, I expect low and middle tiers to bottom in February '09. High tier will take longer.
P.S. closings through 5/22 suggest that the city is on track to shed at least another 1% off the peak in May. So much for the bottom. (That's another 4 billion dollars of "paper wealth" lost in one county, in one month)
Is it, or is it not?
* April is not a likely month to mark the bottom. Barring major interest rate moves or bailouts (which we did not have), typical seasonal patterns suggest that the bottom (as observed by a HPI) should occur in January or February. If you use Case-Shiller, you'll see it in February or March because its results are delayed by one month. Seasonal bounce can temporarily override the correction. It did that every year during the 90's bust. But, if it does not manage to do that till April, the correction is probably not over.
90's San Diego "false bottoms":
1993: May
1994: February
1995: February
True bottom: February of 1996
* Housing price uptick is only seen in median prices, not in this HPI. Median prices are notorious for their high noise levels. Whether we see an uptick in May, remains to be seen.
* Low and middle tier are close to fundamentals, but we're going through never-before-seen numbers of foreclosures (dwarfing anything recorded during the 90's bust) and those are likely to maintain some downward pressure on prices.
* High tier is nowhere close to fundamentals and '04-'05 vintage neg-ams are yet to reset to fully amortizing.
Based on current dynamics, I expect low and middle tiers to bottom in February '09. High tier will take longer.
P.S. closings through 5/22 suggest that the city is on track to shed at least another 1% off the peak in May. So much for the bottom. (That's another 4 billion dollars of "paper wealth" lost in one county, in one month)
Friday, May 2, 2008
Million dollar houses
There are approximately 50,000 houses in San Diego County with current market values above 1 million dollars.
As you could imagine, San Diego multimillionaires tend to stick together. Half of these 50,000 are in only 5 zip codes.
92037 (La Jolla): 14%
92130 (Carmel Valley): 12%
92067 (Rancho Santa Fe): 9%
92024 (Encinitas): 8%
92014 (Del Mar): 6%
92118 (Coronado): 6%
92064 (Poway): 5%
92009 (Carlsbad SE): 5%
92127 (Santaluz, 4S, Del Sur): 4%
92011 (Carlsbad SW): 3%
As you could imagine, San Diego multimillionaires tend to stick together. Half of these 50,000 are in only 5 zip codes.
92037 (La Jolla): 14%
92130 (Carmel Valley): 12%
92067 (Rancho Santa Fe): 9%
92024 (Encinitas): 8%
92014 (Del Mar): 6%
92118 (Coronado): 6%
92064 (Poway): 5%
92009 (Carlsbad SE): 5%
92127 (Santaluz, 4S, Del Sur): 4%
92011 (Carlsbad SW): 3%
Thursday, May 1, 2008
April
City average: 30.2% off the peak, August '03
Top tier: 12.6% off the peak, May '04
Middle tier: 31.4% off the peak, July '03
Bottom tier: 37.4% off the peak, April '03
Definite signs of (temporary?) stabilization everywhere except the very low end. Clairemont/Mira Mesa and Carmel Valley/4S are up slightly. Eastlake/Otay Ranch only lost 0.5%. Still a lot of hurt in deep subprime land: southwest (west Chula Vista, Imperial Beach) and 54-94 corridor are down another 3% month to month. A typical house bought in 54-94 corridor (National City, Logan Heights, Encanto) at the peak in April '06 is now worth less than 58% of its purchase price.
This is how house values are distributed
For example, a house valued at $600,000 is worth more than 75% of all detached houses in the county.
Note: this graph does not account for new construction (it assumes that all houses have been around forever).
Top tier: 12.6% off the peak, May '04
Middle tier: 31.4% off the peak, July '03
Bottom tier: 37.4% off the peak, April '03
Definite signs of (temporary?) stabilization everywhere except the very low end. Clairemont/Mira Mesa and Carmel Valley/4S are up slightly. Eastlake/Otay Ranch only lost 0.5%. Still a lot of hurt in deep subprime land: southwest (west Chula Vista, Imperial Beach) and 54-94 corridor are down another 3% month to month. A typical house bought in 54-94 corridor (National City, Logan Heights, Encanto) at the peak in April '06 is now worth less than 58% of its purchase price.
This is how house values are distributed
For example, a house valued at $600,000 is worth more than 75% of all detached houses in the county.
Note: this graph does not account for new construction (it assumes that all houses have been around forever).
Tuesday, April 1, 2008
March
City average 28.3% off the peak (back to October '03). February city average was
revised to 26.9% off.
Top tier ("fortress"): 11.8% off the peak, May '04
Middle tier ("boundary"): 29.8% off the peak, September '03
Bottom tier ("subprime land"): 35.3% off the peak, July '03
There are some indications of slowing aka "spring bounce", some of it is seasonal, some of it is the consequence of low interest rates of second half of January.
revised to 26.9% off.
Top tier ("fortress"): 11.8% off the peak, May '04
Middle tier ("boundary"): 29.8% off the peak, September '03
Bottom tier ("subprime land"): 35.3% off the peak, July '03
There are some indications of slowing aka "spring bounce", some of it is seasonal, some of it is the consequence of low interest rates of second half of January.
Saturday, March 1, 2008
February HPI
City average 27.3% off the peak.
Here's a different perspective.
Definitions of "fortress" and "undesirables" are fairly subjective. "Fortress" areas (25% of the housing stock) are generally expensive, have low crime levels and good schools. "Undesirables" (also around 25%) have problems with crime, poor schools, or both. For the purposes of constructing this graph, they are defined as follows.
Maps are generated using http://www.usnaviguide.com/zipunion.htm.
Here's a different perspective.
Definitions of "fortress" and "undesirables" are fairly subjective. "Fortress" areas (25% of the housing stock) are generally expensive, have low crime levels and good schools. "Undesirables" (also around 25%) have problems with crime, poor schools, or both. For the purposes of constructing this graph, they are defined as follows.
Maps are generated using http://www.usnaviguide.com/zipunion.htm.
Wednesday, February 6, 2008
Constructing a HPI
Like Case-Shiller or OFHEO HPI, this index is based on repeat sales of same properties. I restrict my attention to detached single-family houses and exclude condos. (Condo prices may have very different dynamics from SFRs in the same area, and it's less work to do only SFRs. Also, Case-Shiller only includes single-family housing, and I like to be able to check against their numbers.)
First we need some resale data. MLS has everything we want, but one needs to be a realtor to access it directly. (Any friendly realtors out there?) Fortunately for San Diego County, there's a site with a lot of sale records going back to 1997:
http://users.ixpres.com/~gtriphan
It isn't perfect (incomplete, lots of misspellings), but it's better than nothing. We supplement it with another site where we can get recent closings:
http://www.sdlookup.com
We can write a script that pulls web pages from those two sites and parses them to extract resale entries. Extracted data needs to be scrubbed a little bit. For example, the same property might be present in our data as "123 10th st", "123 tenth st" and "123 tenth street".
Next step is to identify resale pairs. For every unique property for which two or more resale entries are recorded, we sort them by date. Our data probably still contains a lot of junk (incorrectly entered sales prices, non-arms-length transactions), so it is a good idea to apply some filters. We also want to exclude "flips" or any other resale pairs involving a substantial change in the condition of the property. A possible set of filters:
* Exclude all resale pairs separated by less than 6 months
* Exclude all resale pairs where either one of two resale prices is absurdly low (<$2000) or extremely high (>$10,000,000)
* Exclude all resale pairs with price change of more than 2x in either direction in less than 12 months
* Exclude all resale pairs with average annual appreciation/depreciation rate of 50%/year
The specific choice of filters won't significantly affect the outcome, but proper filters will reduce the noise level in our HPI.
Resale pairs are grouped into areas. The idea is that homes in the same area will experience similar rates of appreciation and depreciation. Also, running the algorithm on individual zip codes typically results in too much noise because of insufficient amount of data. The more data we have, the better the quality of our final HPI. This step requires some knowledge of different neighborhoods. For example, it makes sense to put zip codes 91913 and 91915 into the same group, but their neighboring 91911 has very different demographics and housing stock and it may evolve differently.
Finally we take resale pairs in each area and try to approximate them with a single index using a modification of the weighted least squares algorithm.
I might expand on this later; for now, here's a code snippet:
First we need some resale data. MLS has everything we want, but one needs to be a realtor to access it directly. (Any friendly realtors out there?) Fortunately for San Diego County, there's a site with a lot of sale records going back to 1997:
http://users.ixpres.com/~gtriphan
It isn't perfect (incomplete, lots of misspellings), but it's better than nothing. We supplement it with another site where we can get recent closings:
http://www.sdlookup.com
We can write a script that pulls web pages from those two sites and parses them to extract resale entries. Extracted data needs to be scrubbed a little bit. For example, the same property might be present in our data as "123 10th st", "123 tenth st" and "123 tenth street".
Next step is to identify resale pairs. For every unique property for which two or more resale entries are recorded, we sort them by date. Our data probably still contains a lot of junk (incorrectly entered sales prices, non-arms-length transactions), so it is a good idea to apply some filters. We also want to exclude "flips" or any other resale pairs involving a substantial change in the condition of the property. A possible set of filters:
* Exclude all resale pairs separated by less than 6 months
* Exclude all resale pairs where either one of two resale prices is absurdly low (<$2000) or extremely high (>$10,000,000)
* Exclude all resale pairs with price change of more than 2x in either direction in less than 12 months
* Exclude all resale pairs with average annual appreciation/depreciation rate of 50%/year
The specific choice of filters won't significantly affect the outcome, but proper filters will reduce the noise level in our HPI.
Resale pairs are grouped into areas. The idea is that homes in the same area will experience similar rates of appreciation and depreciation. Also, running the algorithm on individual zip codes typically results in too much noise because of insufficient amount of data. The more data we have, the better the quality of our final HPI. This step requires some knowledge of different neighborhoods. For example, it makes sense to put zip codes 91913 and 91915 into the same group, but their neighboring 91911 has very different demographics and housing stock and it may evolve differently.
Finally we take resale pairs in each area and try to approximate them with a single index using a modification of the weighted least squares algorithm.
I might expand on this later; for now, here's a code snippet:
// "vn1" and "vn2" - dates of first and second sale
// "price1" and "price2" - recorded prices
// "nmax" - the number of months for which we construct the index (if we're working with 1999-2007, nmax=108)
// "sol" - logarithm of our final HPI
int N = nmax-1;
double* matrix = new double[N*N];
double* rh = new double[N];
double* sol = new double[N+1];
sol[N] = 0;
memset(matrix, 0, N*N*sizeof(double));
memset(rh, 0, N*sizeof(double));
for(i=0; i<nPairs; i++)
{
double weight = 1;
// use lower weights when sales are separated by long periods of time
weight *= 1 - 0.4 * abs(vn2[i]-vn1[i]) / 120.;
double ratio = log((double)prices2[i] / (double)prices1[i]);
// make the index smoother by creating extra "copies" of every resale, shifted one month back and one month forward
for(int delta=-1; delta<=1; delta++)
{
j = vn1[i] + delta;
int k = vn2[i] + delta;
if(j<0 || k<0)
continue;
if(j>N || k>N)
continue;
if(vn1[i]==N-1 && j==N)
continue;
if(vn2[i]==N-1 && k==N)
continue;
matrix[j+j*N] += weight;
if(k<N)
matrix[k+j*N] -= weight;
rh[j] -= ratio*weight;
if(k<N)
{
matrix[k+k*N] += weight;
if(j<N)
matrix[j+k*N] -= weight;
rh[k] += ratio*weight;
}
}
}
/** Solve the system of equations **/
for(j=0; j<N; j++)
{
double x = matrix[j + j * N];
if(fabs(x) < 1e-8)
{
for(i=0; i<N; i++)
matrix[i + j*N] = 0;
x = matrix[j + j*N] = 1;
rh[j] = 0;
}
for(i=0; i<N; i++)
matrix[i+j*N] /= x;
rh[j] /= x;
for(i=j+1; i<N; i++)
{
int k;
double mul = matrix[j+i*N];
for(k=0; k<N; k++)
matrix[k+i*N] -= mul * matrix[k+j*N];
rh[i] -= mul*rh[j];
}
}
for(j=N-1; j>=0; j--)
{
sol[j] = rh[j] / matrix[j+j*N];
for(i=j-1; i>=0; i--)
{
rh[i] -= sol[j] * matrix[j+i*N];
matrix[j+i*N] = 0;
}
}
Saturday, February 2, 2008
January '08 - preliminary
Saturday, January 26, 2008
New developments
This is a look at three major new developments in the county: 4S Ranch (with its older sibling Bernardo Springs), San Elijo Hills/Old Creek Ranch, and Eastlake/Otay Ranch.
Take these graphs with a grain of salt. There's a lot of noise due to insufficient data: less than 200 resale pairs are available for 4S Ranch and San Elijo Hills. I'm surprised that graphs came out as nice as they are.
Notice the periodic structure of the 4S Ranch chart. Seasonal variations are fairly common but they are far stronger in 4S than anywhere else. (Has something to do with Poway school district, maybe?)
Take these graphs with a grain of salt. There's a lot of noise due to insufficient data: less than 200 resale pairs are available for 4S Ranch and San Elijo Hills. I'm surprised that graphs came out as nice as they are.
Notice the periodic structure of the 4S Ranch chart. Seasonal variations are fairly common but they are far stronger in 4S than anywhere else. (Has something to do with Poway school district, maybe?)
Tuesday, January 22, 2008
Affordability - 12/2007
"Affordability" is basically an inflation-adjusted mortgage payment for a median house, normalized so that July 2000 = 100%. To measure inflation, I use CPI. (Actual incomes are good too, but incomes in San Diego county got artificially inflated due to all the bubble money flowing in.)
Some assumptions that go into this model:
* Average conventional mortgage rate as published by St Louis Fed
* Coast and RB/RP/4S etc. require jumbo mortgages (1% premium over conventional rate, starting in August 2007)
* Down payment equal to 20% of July 2000 median house price, adjusted for inflation
* To simplify things, it's assumed that the entire balance is financed using the aforementioned rate (although in reality, the amount borrowed would be above 80% of purchase price, so a 2nd mortgage with higher interest rate would be required)
* Property tax equal to 1% of purchase price
Things are rapidly normalizing, partly because of falling prices, partly because of low mortgage interest rates.
Sunday, January 20, 2008
December 2007 data
Countywide, as of December, we're 22.5% off the peak:
My numbers are 3 months ahead of Case-Shiller. The official Case-Shiller for October is 13.1% off the peak and it really reflects the situation as of September, because their October numbers are averages of August through October. We had a 9% decline in 3 months.
Not every area is 22% down, though:
This plot will help explain why. It shows relative appreciation vs. July of 2000.
This is not to say that "Coast" and other upscale areas don't need to fall. Of the two causes of the housing bubble (speculative/investment buying and subprime) they were spared one and it kept prices relatively sane. Yet, a 15-20% drop is warranted just to get prices back to 2003 levels (most other areas are already closing in on 2003 levels fast). Besides, jumbo mortgages are more expensive and harder to come by than in 2003.
Area definitions:
"Southeast": Bonita (91902), Eastlake, Otay Ranch (91913, 91914, 91915)
"Southwest": old Chula Vista (91910, 91911), Nestor (92154), Imperial Beach (91932), San Ysidro (92173)
"Coast": all coastal towns from Coronado to Oceanside, UTC, Pacific Beach, Ocean Beach.
My numbers are 3 months ahead of Case-Shiller. The official Case-Shiller for October is 13.1% off the peak and it really reflects the situation as of September, because their October numbers are averages of August through October. We had a 9% decline in 3 months.
Not every area is 22% down, though:
This plot will help explain why. It shows relative appreciation vs. July of 2000.
This is not to say that "Coast" and other upscale areas don't need to fall. Of the two causes of the housing bubble (speculative/investment buying and subprime) they were spared one and it kept prices relatively sane. Yet, a 15-20% drop is warranted just to get prices back to 2003 levels (most other areas are already closing in on 2003 levels fast). Besides, jumbo mortgages are more expensive and harder to come by than in 2003.
Area definitions:
"Southeast": Bonita (91902), Eastlake, Otay Ranch (91913, 91914, 91915)
"Southwest": old Chula Vista (91910, 91911), Nestor (92154), Imperial Beach (91932), San Ysidro (92173)
"Coast": all coastal towns from Coronado to Oceanside, UTC, Pacific Beach, Ocean Beach.
Subscribe to:
Posts (Atom)