Academic Articles Economics - General

The Effect of Health Shocks and Aging on Asset Allocation

If one does not include Social Security and defined benefit assets, households with heads aged 65-74 have net assets of $190,100.  Households with heads aged 75 and above have net assets of $163,100.  How do the elderly allocate their assets?  Does their portfolio choice change over time?  Does it change in response to health shocks?  

These are the questions which Coile and Milligan (2006) seek to answer.  In their research, they found that their was little evidence that the elderly draw down their assets substantially during retirement [Guiso, Haliassos, Jappelli (2002); Borsch-Spuppan (2003)].  Milligan (2005) also found that in Canada, most households do not sell their houses or vehicles until late in life.  One reason for the paucity of home sales may be due to transaction costs.  For the elderly, the largest transaction cost may be the psychic cost of having to uproot themselves from their familiar abode. 


Coile and Milligan use data from the Health and Retirement Survey between 1992 and 2002.  Their econometric specifications are very simple:

  • Asset holdingsit = b0 +b1ageit +b 2Xit +g t +e it


The asset holdings (i) include five categories: stocks/bonds, a house, a vehicle, bank accounts/CDs, and business/real estate.  Here the aim is to examining trends as individuals age.  The authors later include dummies for 3 shocks: 1) having a spouse die, 2) being diagnosed with a chronic illness, and 3) experiencing an acute event (such as a heart attack, stroke, etc.).  Lags and leads of these shocks are also included.  Finally, cohort based fixed effects by birth year are used as well as household fixed effects.  As authors are wise to explain the following:

In comparing the linear age, cohort fixed effect, and household fixed effect specifications, the usual trade-offs apply – the specifications with cohort and household fixed effects likely do a better job of controlling for unobservable heterogeneity, but there is a risk of being left with too little variation to estimate statistically significant relationships.


The authors find that home ownership is fairly steady at 80% until age 80.  After that home ownership drops off to 54% at age 90.  Similar results are found for vehicle ownership.  The authors found that becoming a widow is a strong predictor of selling one’s principal residence.  This results has been found in other economics studies, but what is new is that the widow shock also decreases ownership of vehicles, businesses and real estate, while increasing the share of assets held in bank accounts and CDs.  The authors also find similar results for the health shocks, but the findings are not as robust.  Negative health shocks vary in magnitude, and measuring the precise date of ‘poor health’ may be difficult if one’s health was already deteriorating before the ‘shock.’  We see a pattern of movement from more to less risky investments as one nears the end of one’s life.  The pattern is magnified when a negative health shock or death of a spouse occurs.  Both shocks may lead to increased spending (on health products or substitutes for household labor), and thus the need to maintain a minimum level of non-volatile income through CDs.  One issue I believe the authors missed is that a family’s wealth may greatly affect its asset allocation.  Using an instrument of initial wealth, and creating dummies for different wealth percentiles may have lead to more accurate analysis of what is causing behavioral changes. 

Coile, Milligan (2006) “How Housing Portfolios Evolve after Retirement: The Effect of Aging and Health Shocks,” NBER Working Paper #12391.