Solve the model allowing for heterogeneous beta types, where beta types have differential education weight. So this involves solving the model for education/beta subgroups. The weighted average of results from different beta/edu files produces mixture model results for the economy.

Marginal utility drop in one period to model consumption drop during lock-down.

Visualize allocation results and algorithm.

• Possibly change matlab globals to map (2020-06-17 09:28)
• Y function, C function, possibly functionalize rather than retype (2020-06-17 10:33)
• main.m-L248: given new a2 resolve policy until gov budget clears (2020-06-17 10:32)
• V_working_proxy-L22: unable to approximate the value of working or unemployed at the highest grid asset point (2020-06-17 11:11)
• V_working_proxy-L51: solved for a_aux, V(a_aux_bounds) should already be known (2020-06-17 11:21)
• pi_unemp_calibration: deriving joint from conditional distributions, without restrictions, there are infinite number of solutions, this might not be unique solutions now (2020-06-17 11:48)

Tax Issues

• Should kid count impact tax rate (2020-06-17 09:39)
• in mana from heaven case, we have two tax rate, a2_2020 to clear 2020 costs of covid checks of SS costs, and a "stationary" a2 for Vss. Now we go from a2_2020 to a2_stationary in 2020 right away, which means a2_stationary will not actually clear gov budget because there will be less savings than we used to clear a2. SO need to adjust transition path. (2020-08-08 08:51)

Aggregation Issues

• At the moment, the "agent" in the model is the household head. The agent, however, might be married and might have kids. So technically the size of the population is larger. That might be something we need to keep in mind when we calibrate the model since for example total factor productivity, theta, is set such that income per capita is equal to 1. What we count as population in the model is not really correct since we're not accounting for spouses and children. (2020-06-25 14:50)
• Check if we have to adjust savings to account for population growth Since it's not GE, we might be correct. See Conesa et. al. 2020. (2020-06-28 17:52)

  But just based on the literature I'm used to, it still tends to show up in the resource constraint through multiplication with AP. But I just can't see exactly how it should look like since I'm not used to thinking about non-GE environment. Phi_true already has g_n built in. That's something we need because otherwise there will be way too many old people relative to young people, so the average tax burden (and therefore consumption) will be way too low compared to the data. So that's taken care of when we compute the population by age. But still feel like it should appear in the resource constraint.

Shock Process

• Right now there are two income shocks before retirement, and only spousal shock matters after retirement. Do we want to have both shocks to disappear after retirement? (2020-07-14 13:14)

Unemployment

• unemployment for husband and wife are fully correlated, should they be? (2020-08-07 15:49)

Memory usage not very efficient.

1. Memory usage very high as shocks increase.
2. Memory usage ver high during initialization of parallelization, leading to limited available cores.

• Does model stationary distribution match with 2020 observed (2020-06-17 09:10)
• Currently, everything in the model is normalized to real income per capita, which was about 60,000 last year. Since we have households in the model, we can compare average household income in the model and the data. Average household income was about 90,000 last year. I'm much more used to thinking about individuals. So I think that means the calibration target for theta is wrong. Right now, we're calibrating theta such that aggregate income (so integrating over interest earnings, labor earnings, social security benefits, and spousal income) over aggregate population equals 1. Where a value of 1 corresponds to 58,000 dollars. Hmm, actually, I'm not sure. If population in the model is the entire population in the economy, then that does indeed correspond to income per capita. I guess the problem is that the population doesn't really refer to the population since we're only counting household heads.

Planner Objective Issues

• V_U and V_W, as well as C_U and C_W are both computed by linear interpolation. To be fully consistent, one might want to "You sort of want to carry over the approximation from C to V" (2020-07-07 10:44).

Given COVID welfare checks, solve for distribution of asset for the next period under MIT shock.

There is a prior round of welfare checks in year X which experienced some MIT shock. In year X+1, there is another MIT shock, however, the distribution of households entering this period is different due to the MIT shock last period as well as the checks received last period (only checks matter if manna from heave and last period no income loss)

Distributional code speed main speed stumbling block

Three separate things:

1. weighted average of individual MPC files for joint MPC files
2. consider the biden-check schedule and related policies
3. weighted average allowing for inequality aversion weighting to work with inequality aversion results.