``The Federal Student Loan Program: Quantitative Implications 
  for College Enrollment and Default Rates'' (manuscript # 
  RED-07-196)

  Description of codes

There are 21 codes to solve the model and deliver the results in 
the paper. They are grouped in 4 categories: 1) 4 codes that 
deliver the main results in the paper and the distribution of 
initial characteristics. Note that this is the last step in the 
algorithm. 2) 3 codes that are used to solve for the no college 
path, which in the end deliver the earnings profiles for the no 
college group to be used in the codes in part 1. 3) 11 codes that 
are used to solve for the college path, which deliver the 
earnings profiles for the college group to be used in the codes 
in part 1. 4) 3 auxiliary codes used in simulations. In addition, 
there are 3 files with earnings statistics from data. 

1 Codes for the initial distribution and the main results in the 
  paper

This part describes the use of the following codes: 
main_enrollment.m, distribfind_step3_colnocol.m, 
simulation_default.m, simulation_earnings.m, 
simulation_welfare.m. 

1. All the codes in this section are run from main_enrollment.m. 
  This code also solves for the optimal enrollment decision.

2. The code distribfind_step3_colnocol.m delivers the joint 
  distribution of ability and human capital together with the fit 
  of the model relative to the data earnings statistics presented 
  in the paper in Section 3.3. This represents step 3 in the 
  computational algorithm described in the appendix. This code 
  uses simulated earnings profiles for agents in the model that 
  are delivered using codes on both the college and the no 
  college paths presented separately below. It also uses the 
  earnings statistics (mean, skewness, and gini) from the CPS 
  data for high school graduates, data_ALL_CPS2.mat (attached) 
  and alos by education groups, data_CG_CPS2.mat and 
  data_NC_CPS2.mat (also attached). The code needs the objfct.m 
  file in the auxiliary part.

3. The code simulation_welfare.m delivers most of the results in 
  the paper: college enrollment percentages by initial 
  characteristics (human capital, ability, and assets) in the 
  benchmark economy, averages of human capital and ability by 
  education groups (Section 4.1.2), college enrollment and 
  default rates in the benchmark economy and in the two policy 
  environments (Sections 4.2 and 4.3.1), life-cycle earnings of 
  defaulters versus nondefaulters in the benchmark economy 
  (Section 4.2), consequences for the government budgets (Section 
  4.3.2), and allocational consequences by initial 
  characteristics in the flexible repayment system, which are 
  discussed in Section 4.3.3.

4. In addition, the code simulation_earnings.m delivers earnings 
  profiles for the two education groups in the benchmark economy 
  and in the flexible repayment system (Section 4.3.1).

The simulation codes in this section use the auxiliary codes 
listed below in part 4. To run this part, intermediary steps are 
needed (steps 1 and 2 in the computation algorithm in the 
appendix). The codes needed for these two steps are described 
below for each path.

2 No college path codes

This part describes the use of the following 3 codes: main_nc.m, 
hchoice_step1_nc, and earnings_step2_nc.m.

1. All the codes in this section are run from main_nc.m. 

2. The code hchoice_step1_nc.m solves for the optimal decision 
  rules on the no college path on the grid of learning ability, 
  initial human capital, and initial assets (a,h,x). This is step 
  1a) in the computation algorithm in the appendix.

3. The file earnings_step2_nc.m computes life-cycle profiles of 
  human capital, hours, and earnings from the grid points for an 
  extracted number of points in the initial grids. This is part 
  of the step 2 in the computation algorithm in the appendix. 
  This code delivers the earnings for the no college group used 
  in distribfind_step3_colnocol.m and main_enrollment.m.

3 College path codes

This part describes the use of the following 11 codes: main_cg.m, 
hchoice_step1_cgpp.m, hchoice_step1_cg.m, DEF.m, 
hchoice_step1_cg_bench.m, IC.m, SC.m, hchoice_step1_cgchoice, 
dhchoice_step1_5.m, earnings_step2_cg_bench.m, and 
earnings_step2_cg_choice.m.

1. All the codes in this section are run from main_cg.m (there is 
  one exception noted below).

2. The codes hchoice_step1_cgpp.m, hchoice_step1_cg.m, DEF.m, 
  hchoice_step1_cg_bench.m, IC.m, SC.m, hchoice_step1_cgchoice.m 
  solve for the optimal decision rules on the college path on the 
  grid of learning ability, initial human capital, and initial 
  assets (a,h,x). This is step 1b) in the computation algorithm 
  in the appendix, which involves several sub steps:

  (a) for post-payment I use hchoice_step1_cgpp.m 

  (b) for payment periods during post-college I solve for each 
    payment option available: no consolidation 
    (hchoice_step1_cg.m), default (DEF.m), income consolidation 
    (IC.m), standard consolidation (SC.m). 

  (c) given decision rules for each payment path, I aggregate the 
    results and choose the optimal payment as long as the agent 
    does not choose to change from the standard fluctuating 
    payments status using hchoice_step1_cg_bench.m for the 
    benchmark economy (and relaxation eligibility system) and 
    using hchoice_step1_cgchoice.m for the flexible repayment 
    system.

  (d) for the college period, optimal choices are computed in 
    hchoice_step1_cg_bench.m and hchoice_step1_cgchoice.m, except 
    for the first period, which is done separately in 
    dhchoice_step1_5.m This code also computes the loan amount 
    given for the grid points. Note that this can be adjusted to 
    deliver the debt and the relevant choices associated with 
    that debt level, in all three environments: the benchmark 
    economy, the relaxation eligibility system, and the flexible 
    repayment system (instructions on this are included in the 
    file). Also, this code runs independently (i.e. not from 
    main_cg.m).

  • An important observation is that all these codes can be run 
    in parallel for a restricted set of ability levels (and time 
    of switch to another payment path) to economize on time.

3. The files earnings_step2_cg_bench.m and 
  earnings_step2_cg_choice.m compute life-cycle profiles of human 
  capital, hours, and earnings from the grid points for an 
  extracted number of points in the initial grids for the 
  benchmark economy (and relaxation eligibility system) and for 
  the flexible repayment system, respectively. This is part of 
  the step 2 in the computation algorithm in the appendix. This 
  code delivers the earnings for the college group used in 
  distribfind_step3_colnocol.m and main_enrollment.m.

4 Auxiliary codes

This part refers to 

1. Markov.m and chooserandom.m, which are used in the simulations 
  codes in part 1) and 

2. objfct.m, which is used in finding the distribution of (a,h), 
  code distribfind_step3_colnocol.m in part 1).