Housing is typically the greatest investment, and the most valuable asset, of any household. Overall, it dominates the household portfolios and is crucial in the accumulation of wealth over time. Since housing assets can serve as collateral, people are granted large mortgages, and thus even modest returns yield great wealth boosts due to the sheer magnitudes of the investments. Naturally, owning a home also produces housing benefits of great value to the owner-occupier. Everyone needs a place to live and owning a home secures a steady flow of dwelling services that would otherwise have had to be bought in the rental market. It is no surprise then that Norwegian households' asset portfolios are undiversified: using data from the Income Distribution Survey 2002 from Statistics Norway, we find that the average portfolio has a housing-to-wealth ratio of 1.175, a stocks-to-wealth ratio of 0.089, and a debt-to-wealth ratio of 0.264. These figures mirror findings in similar surveys of other economies. Breaking the average holdings down to age cohorts shows that young households hold much more housing and debt relative to their wealth than older households. It is also evident that wealth accumulates a great deal over the life-cycle: when population average wealth is normalised to unity, the cohort relative wealth goes from 0.074 for the youngest households, via 2.233 for 55-66 year olds, and finally down to 1.177 for households with head older than 80. The absence of equity holdings are further notable throughout the study: even the portfolios of the wealthiest quartile contain only 10% stocks. Our key motivation is then the apparent question; is this behaviour optimal and can the allocations be rationalised by formal models? If not, what are then the theoretical recommendations? For our model experiments, the return on risk-free holdings (i.e., bank deposits, loans and bonds) is estimated by an average of Norwegian real bond rates over 1992-2006 less 28% tax, and amounts to a rate of 0.03 annually. Stock return is set to the price appreciation (less 28% tax) plus the dividend rate on the Oslo stock exchange, less inflation, over the same time period. Housing return is similar as it also consists of a capital gain component and a dividend stream. While the former is easily observed by the house price index from Statistics Norway, the latter is not readily available due to lack of observations. However, with justification in rental data and in theory of the benefit to the owner-occupier, we settle on the before-tax, real risk-free rate as proxy for housing dividend. We further elect to follow advice from the literature on the riskiness of single-home housing investments, and set the standard deviation equal to 60% of that in the stock market. The latter is found be 0.2 annually for the 1992-2006 period. The reason for not simply using the risk exhibited by the house price index as our measure is that this aggregates market transactions and thus cannot capture the risk of buying a single house. We find mean rates of returns from housing and stocks of 0.11 and 0.188 respectively, with a correlation of 0.33. Real prices on stocks and housing have appreciated tremendously since the early 1990s. In fact, real prices were fairly stable -- or even decreasing -- over most of the 20th century, before going through a boom-bust period in the mid-to-late eighties, and then literally taking off around 1993. Since then, real prices have grown by roughly 490% and 260% respectively, and the returns to equity- and house owners have clearly been enormous over these years. Our first attempt at explaining or rationalising the observed household behaviour mentioned above is the employment of a static mean-variance model of portfolio choice. We assume that the investor only cares about the expected return and variance of the portfolio and that the objective is to minimise this variance given a requirement on the expected return (this is of course equivalent to maximising expected return given some risk tolerance). There are two risky assets, housing and stocks, and one risk-free asset, simply called "risk-free." The two-fund separation theorem ensures that any investor household will choose a combination of the risk-free asset and the market portfolio. While the latter is determined by the properties of the assets available to all investors, each investor will put a weight on the risky- and the risk-free portfolio according to his or her level of risk aversion. The solution to such a setup will then be a set of shares of housing, stocks and risk-free. With return estimates based on observed asset performances over 1992-2006 we find that the holdings of stocks and housing should be virtually equal in magnitude (the market portfolio composition of risky assets). The less risk averse the household is, the more stocks, housing and debt should it take on. More risk averse households on the other hand, should hold less of the risky assets and positive amounts of the risk-free asset (i.e., bank deposits or bond holdings). Since this result is very far from the observed holdings, we seek an alternative solution by using different return estimates. Reliable data on equity dividend and on house- and stock prices are available from 1966, so we collect asset properties for the 1966-1991 period and calculate a new set of estimates. The resulting portfolio model solution is uplifting: the framework now prescribes far less stocks relative to housing, which we know is closer to actual allocations. For example, with a risk aversion coefficient equal to 3 we find an "optimal" housing share of 1.129, a share of stocks of 0.484, and a residual risk-free share of -0.613. This is not very far from the observed, average household portfolio barring the weight on stocks (which is much lower in the data). Further, there seems to be a clear relationship between age and level of risk aversion. That is, for both sets of results, the high-risk aversion portfolios match the observed holdings of older households better, while the model portfolios with lower risk aversion coefficients are better fits with younger households' allocations. We have mentioned that younger households have the largest shares of housing in their overall portfolio, and that the share tends to decline with age. If we now, as an extension of the applied portfolio model, assume that the amount of housing is fixed and determined by the households' demand for housing service consumption, it may be interesting to find the optimal portfolios given the housing share of each cohort. That is, what are the optimal shares of stocks and risk-free when the household is already equipped with a certain expected return and risk from the housing holdings? Unfortunately, our analytical results make no sense in this case due to the absence of a credit constraint: the model prescribes outrageously high portfolio shares of stocks and debt. However, by imposing a limitation on how much the households are allowed to borrow ("no more than minus the housing holdings") in a simple numerical procedure in Excel, we obtain far more reasonable results. But, with the 1992-2006 estimates the model recommends an equity share (which is never higher than 0.12 in the data) between 0.75 and 1 for all cohorts and a debt share consistently greater than that in the data. When we instead assume that the 1966-1991 estimates hold, the match with data is almost perfect: low portfolio shares of stocks throughout and a share of risk-free virtually in sync with the Norwegian observations from 2002. It thus seems like people either disregard the tremendous returns of the nineties and early noughties as extraordinary and unsustainable, or that they enjoy even greater utility from housing than what we have accounted for in our efforts. In any event, the older return estimates are better able to explain the observed 2002 holdings than estimates based on asset returns seen in the years around the observation itself. Our second main exercise is the employment of a dynamic life-cycle model. This approach allows a far richer economic environment to be constructed for the household as we now can consider intertemporal investment-consumption choice and explicitly model the consequences of decisions. It is then assumed that the household receives utility from regular consumption and from housing consumption, and that the objective is, at all stages of life, to maximise the present value of the remaining lifetime utility. While there are still two risky assets and one risk-free asset, we are now adding a stream of labour income which is to be optimally allocated between investment/savings and immediate consumption. As the lifetime optimisation problem is virtually impossible to solve analytically, we resort to the well-known numerical method of value function iterations. This procedure basically entails approximating an unknown function (the present value of remaining lifetime utility -- the value function) by identifying the values of the choice variables that yield the greatest lifetime utility at any point in time, and for all possible past investment decisions. Using this function we next find consumption- and investment rules that prescribe optimal actions given any previous set of allocations and asset returns, and thus solve the household's problem. It is further assumed that the representative household begins life at age 20, dies at age 80, and that every period in the model corresponds to five years. For computational tractability we abstract from income/human capital uncertainty and simply set the household's income equal to the (normalised) average Norwegian income in 2002 for every period/age. Since the solution is that of a representative household, we simulate the "lives" of 400 individual households, find the average allocations and interpret these as our solution. With the 1992-2006 asset return estimates we find that the household should invest quite heavily in housing from the get-go by taking on the maximum amount of debt (relative to the housing investment), while equity investment and regular consumption are chiefly postponed till age 35. From then on however, consumption levels grow tremendously and lie far above labour income throughout. While equity holdings also grow over the life-cycle and stabilise nicely between housing holdings and labour income, debts are responsibly paid down by the simulated households. Moreover, as the stock holdings accumulate and debts are abolished, housing stays fairly constant over time. A possible explanation is that households are using the capital gains from their housing asset to increase consumption, pay down the mortgage and invest in the stock market, rather than reinvesting it back into more housing. This makes perfectly good sense in our model, but may not be too reasonable in the real world. With the 1966-1991 asset return estimates we find mostly the same average profiles except for holdings of stocks and housing being consistently greater than with the 1992-2006 estimates. This sounds a bit curious since the former asset returns are lower than the latter: why would the households invest more now? One rationale is that the amounts invested in various assets must be higher when expected returns are lower in order for the household to attain a comfortable level of precautionary savings, and to be able to push up future consumption. On the other hand, lower returns on savings also means that future consumption is more expensive, thus households should save less and "eat" more. As our model output shows that regular consumption is pushed forward but attains lower overall levels over the life-cycle compared to the output with the 1992-2006 estimates, both above effects are recognised. In the final periods of these exercises, equity investments drop to zero since we assume that there is no value in leaving anything behind. But what if we instead say that the household draws utility, while alive, from bequeathing its end-of-life worth to a younger generation? Solving this modified problem we find, as expected, that the simulated households consume less and save more at the end of their lives. Such a specification is of course much closer to reality since it is commonly assumed that bequest motives exist, and because of the inherent uncertainty of the time of death (which we characteristically have abstracted from in our theoretical escapades). Finally, we compare the performance of the (bequest) model, with the two sets of estimates, to the observed Norwegian life-cycle holdings, surveyed in 2002. We find that the newer estimates allocations lie much closer to the actual holdings than those of the older estimates. This is immediately interesting because the static portfolio model produced the exact opposite conclusion: older estimates were able to better explain observed behaviour than newer ones. Even though predicted stock holdings are still wildly out of tune with observations, the dynamic model solved with the 1992-2006 estimates can explain the broad features of households' behaviour.