Professional Documents
Culture Documents
Support System
361-816-3532
kylewesleytaylor@gmail.com
May 2014
1
Précis
decision support system. The goal is to show that increasing model complexity via
the addition of more variables and a greater quantity of data will yield a more
accurate model. This has been done by creating a simple financial model intended
to break down the overall actual costs of attending a university and the expected
monthly loan payment a graduate from said university can expect to have, as well as
the impact of the loans on their salary. All of the variables in the first model, though
real-world data, are assumed to be static for the life of the model. After analyzing
the results of the first model, a second model is proposed wherein fewer
assumptions are made and fewer variables are assumed to be static. Using
historical data over a period of time, time-trend analysis is used in order to acquire
the best-fit predictive functions for the variables: interest and tuition price. In
addition, more research allows for a better salary approximation based upon major
student loans are also split between federal subsidized loans and federal
unsubsidized loans based upon the percentage of tuition allowed to be taken out for
Upon completion of the second model, a much more reasonable and realistic
and potential model subsets are explored in improving the second model into an
even more complex third model. It is concluded that increasing model complexity
via increasing data quantities for analyzing yields a more accurate result.
2
Table of Contents
Research Question…………………………………………………………………………………….Pg. 7
Conclusions……………………………………………………………………………………………..Pg. 32
References……………………………………………………………………………………………….Pg. 35
Appendix…………………………………………………………………………………………………Pg. 37
3
Tables and Figures
Table 11: Cost, Note Payable, and Interest Accrued During College.............Pg. 25
4
Introduction to Quantitative Analytics
In the age of “Big Data,” the need for a workforce with quantitative skills is in
high demand. A quantitative analyst, or quant, is someone that can sift through
large amounts of data and build a model to accurately foretell an outcome. In the
illiquid) gives rise to people who can not only understand the models, but can also
manage or enhance them. For instance, a relatively new sector in the financial
market in order to build an alpha generation platform. Alpha is a term used in the
standard before the “alpha generation platform”, or model, can be executed – and
(Jay, 2008).
Quants aren’t limited to the financial industry only, but can work in a number
company’s terabytes upon terabytes of data, apply the relevant data to a model, and
see where the company needs to streamline efficiency, cut costs, or allocate funds
more heavily in an area within the company. Quantitative analysts may also work in
5
financial skills (Yates, 2013). However, it isn’t just the complex skillset that make
research question to an intricate and multifarious problem that sets them apart, as
fortunetellers.
6
Research Question
desired answer, and from there will be improved upon by adding more complexity
to build a second model, which hopefully will yield a more accurate answer. The
topic chosen to illustrate this is that of the actual cost of attending college and
student loan debt. The number of 25-year-old Americans with student debt has
risen from 25% in 2003 to 43% in 2012. Outstanding student loan debt topped $1
trillion for the third quarter of 2013, “with balances and delinquency rates rising to
borrowing” (Gage, 2014). In contrast, auto, mortgage, and credit debt have all
declined from their peaks. Student loan debt can have a pretty significant effect on
the economy. According to the Federal Reserve, the burden of student debt can keep
people unable to buy homes or cars, thereby really affecting the whole economic
cycle and making it an extremely relevant issue at this point in time (Gage, 2014).
Thus, I found it pertinent to construct a model in this field that shows what a
prospective student can expect to pay in entirety for their education, as well as the
disposable income they can expect until full loan repayment. The goal is to increase
model complexity so that the second college financial model yields a more
7
Building Model 1
that would show us the amount per month it would take a student to pay back the
loans they took out for their education. The initial input for the model will be 6
for each university (Pay Scale, n.d.), average annual cost of attendance (CNN Money,
n.d.), and the federal student loan interest rate (U.S. Department of Education, n.d.),
Based on these figures, some assumptions will have to be made about the
model. We will assume the student will receive no grants or scholarship, and the
total cost of attendance will be paid entirely by federal unsubsidized student loans
8
with a fixed interest rate of 3.86% compounded daily from the first day of term. It
will also be assumed the student will have the standard repayment plan of 10 years
according to “FinAid” (n.d.). For all universities, it will be supposed that the annual
cost of attendance loan will be disbursed in the fall semester, assumed to start
August 1st. The student will start their career in June following graduation and
With this information, a simple model can now be built in Excel to show how much
the student will be paying in total for their education, as well as the percentage per
month of their disposable income will be spent on repayment the 10 years following
graduation.
In table 2, the total loan amount to be repaid as the student starts their
career is found by determining the sum of all the future value of cash flows using
FV=PV*(1+i)N, where PV is present value, i is the interest rate, and N is years (i.e.
annual cost * (1+3.86%)3.917). In the last cash flow, N=3.917 because the loan was
disbursed in August of the previous year and the first payment will be made in July,
9
Table 2: Total Note Payable and Interest Accrued During College
Total
Average Total Note
Interest
Annual Total Payable
University Accrued
Cost of Degree Cost Upon
During
Attendance Repayment
College
Princeton
University $57,628.00 $230,512 $253,421.29 $22,909.29
Harvard
University $60,240.00 $240,960 $264,907.65 $23,947.65
Dartmouth
College $64,034.00 $256,136 $281,591.91 $25,455.91
Columbia
University $64,949.00 $259,796 $285,615.66 $25,819.66
Yale
University $61,333.00 $245,332 $269,714.16 $24,382.16
University of
California-
Berkeley $32,479.00 $129,916 $142,827.62 $12,911.62
Pennsylvania
State
University $31,517.00 $126,068 $138,597.19 $12,529.19
New York
University $65,303.00 $261,212.00 $287,172.39 $25,960.39
University of
Texas -
Austin $25,587.00 $102,348.00 $112,519.79 $10,171.79
Rutgers
University of
New Jersey $29,512.00 $118,048.00 $129,780.12 $11,732.12
From this it can be seen that the average interest accrued while attending a
private school is $24,745.84, and for the public universities it is $11,836.18. Now
the monthly payment requirement can be calculated in excel by using the payment
function and inputting the annual interest rate (divided by 12 for monthly interest),
the number of payment periods (10 years = 120 months) and the present value of
the loan which is equal to the total amount of loans and interest accrued up to July
1st. As shown in Table 3, this gives you your monthly payment, and by multiplying
10
the monthly payment by 120 months, the final actual cost of attending each
university is found.
Comparing Table 3 with the Table 2, the total interest paid is found by
subtracting the total amount upon full repayment from the actual degree cost. In
every instance the interest paid is greater than if the student had attended one extra
year of college. This can be broken down further by finding the student’s
prospective monthly salary derived from the average starting salary and comparing
it to their average monthly loan payment. This model is now complete and if the
student would like to attend any other university currently not in the model all the
input they would need is the annual cost of attendance and average starting salary
from the university to fully break down their total degree cost, loan payments, and
disposable salary.
11
Table 4: Effect on Salary
Net
Average Monthly % Monthly Disposable
Monthly Loan Salary To Income
University Salary Payment Loan Month Per Year
Princeton
University $4,858.33 $2,548.94 52.47% $2,309.39 $27,712.73
Harvard
University $4,225.00 $2,664.47 63.06% $1,560.53 $18,726.35
Dartmouth
College $4,508.33 $2,832.28 62.82% $1,676.05 $20,112.61
Columbia
University $4,558.33 $2,872.75 63.02% $1,685.58 $20,226.95
Yale University $4,075.00 $2,712.81 66.57% $1,362.19 $16,346.22
University of
California-
Berkeley $4,341.67 $1,436.58 33.09% $2,905.09 $34,861.09
Pennsylvania
State University $4,050.00 $1,394.03 34.42% $2,655.97 $31,871.69
New York
University $4,058.33 $2,888.41 71.17% $1,169.92 $14,039.06
University of
Texas - Austin $4,066.67 $1,131.74 27.83% $2,934.93 $35,219.16
Rutgers
University of
New Jersey $4,041.67 $1,305.34 32.30% $2,736.32 $32,835.89
12
Analysis and Limitations of Model 1
For the private colleges, over half to nearly three-quarters of the student’s
monthly salary will go to repaying their debt for the next 10 years, while for the
public schools over a quarter to slightly over a third of monthly salary is spent on
repayment. According to our data, NYU has the highest cost of attendance but one of
the lowest average starting salaries, yielding 71.17% of monthly salary spent on
their debt, giving them a yearly disposable income of $14,039.06 until they are
approximately 32 years old. According to the U.S. Department of Health & Human
Services (2014), that is only 20.3% above the 2014 poverty threshold of $11,670 for
a single person household. Clearly, the results from the forecasting model show a
question, and show a significant skew between the average starting salary and
actual disposable salary the graduate would be making whilst in repayment of their
debt.
However, it must be seen that there are so many assumptions made for this
model that it is overly simplistic and thus not very accurate. For example, taking the
average starting salary from universities with over a hundred different majors will
substantially distort the results. The average starting salary between a business
major and a drama major, for instance, would in all likeliness be vastly different.
Another noteworthy mistake is the assumption of ceteris paribus, or that all the
13
to assume one time-stamped variable to remain perpetual in the future. Indeed, for
any analyst the only constant they may truly depend on is that of change.
14
Building Model 2
data over a period of time and conducting a time trend analysis will yield the data
set’s best fit forecasting equation (linear, logarithmic, exponential, polynomial, etc.),
or if the data does not fit a time trend, matching it to its best fit distribution (i.e.
mean and standard deviation for future forecasts. It will no longer be assumed that
the cost of attendance will remain constant, but by seeing the historical change in
tuition cost over time it can be taken into account for the student’s four years spent
at college. In addition, the historical federal interest rate on student loans and
percentage change in starting salary based on the student’s major will be analyzed
to improve the forecast. In this particular instance, we will assume the prospective
assumed that the student will bear the full financial burden of college, but that the
student is responsible for tuition only, whilst the expected family contribution (EFC)
will cover the expenses of room and board, meal plans, books, and miscellaneous
fees. Once this data has been analyzed, fit to a distribution or time-trend, the model
may then yield a more accurate cost breakdown and disposable income-until-
repayment.
In our new model, we will look at cost of attendance not as the actual full
cost, but as the historical tuition cost from the years 2002-2010 per school for a
student. Taking the historical data over a number of years will then allow us to see
if there is a time-trend relationship or, if not, to fit it to its best fit distribution for
15
predicting the expected change in tuition cost over the course of the following four
The interest rates on the student loans will be broken up into percentages
Education. The first year of college aid (assuming the student is in the majority and
dependent), the student may not exceed 63.64% of cost of attendance in subsidized
loans with the remaining in unsubsidized loans. For the second year the student
may not exceed 69.23% in subsidized loans, and for the third and fourth year the
By taking the historical federal subsidized and unsubsidized loan rates over
the past 8 years (U.S. Department of Education, n.d.) a table can be put together and
analyzed as such:
The unsubsidized interest rates remain the same for 7 years before dropping nearly
in half in 2013. This can be taken into account in two ways, the 8th year can be
assumed to be an outlier and 6.80% interest can be used in the model, as it is the
mode. However, it can also be interpreted that the unsubsidized rate compared to
the subsidized rate is quite stable, and that if there is such a drastic change in the
last year we must assume it is for a significant reason and will continue to be stable
16
at 3.86% for years to come. In this instance, the second reason seems the more
likely when taking into account the steady decline in the subsidized loan rates.
When analyzing the federal subsidized loan rates there is a clear indication of a
extrapolate a predictive equation based on the best fit to the original data.
The original data is shown on the above time-trend analysis graph. The best-fit
trend is the two-period moving average, however we cannot get an equation for
future predictions with this method, we look at other trendline graphing options
and search for the highest possible R2, or coefficient of determination. The
estimates the real data points, or the “best fit” of the trendline. A coefficient of
17
trendline perfectly fits the real, observed data. The highest possible R2 value that is
achievable with this data set is that of 0.90265 with a polynomial trendline.
Therefore, in predicting the change in interest rates for the next four years we are
given the equation y = 0.0004x2 - 1.4696x + 1,482.1864 with X being equal to the
year and Y being the expected interest rate for that year. It is important to note that
this equation has been rounded off to 4 decimal places, and should be expanded out
in Excel to 30 decimal places to achieve the most accurate result. Thus, our
predicted unsubsidized interest rates for the next four years when the formula is
Based on the above predictions, the percentage of the tuition year loan that
will be paid for by subsidized loans will be fixed at the predicted rates on the date of
disbursement, but will not accrue interest until after the student graduates. The
unsubsidized rates are assumed to be static at 3.86% for the next four years.
Table 7 shows historical change in tuition prices for the universities in our
model (Cost of Tuition for American Colleges and Universities, n.d.) from the years
2002-2010.
18
Table 7: University Tuition From 2002-2010
Tuition
University 2002 2003 2004 2005 2006 2007 2008 2009 2010
Princeton
University $26,160 $27,230 $28,540 $29,910 $31,450 $33,000 $33,000 $34,290 $35,340
Harvard
University $26,019 $27,448 $29,060 $30,620 $32,097 $33,709 $34,998 $36,173 $37,012
Dartmouth
College $26,562 $27,771 $29,145 $30,465 $31,965 $33,501 $35,178 $36,915 $38,679
Columbia
University $26,891 $28,385 $29,788 $31,472 $33,246 $35,166 $37,223 $39,326 $41,316
Yale University $26,100 $27,130 $28,400 $29,820 $31,460 $33,030 $34,530 $35,300 $36,500
University of
California-
Berkeley $15,197 $16,580 $20,068 $23,686 $23,961 $25,338 $26,785 $28,264 $31,022
Pennsylvania
State
University $15,522 $17,610 $18,828 $20,784 $21,744 $22,712 $23,712 $24,940 $25,946
New York
University $25,380 $26,646 $28,496 $30,094 $31,690 $33,420 $35,290 $37,372 $38,765
University of
Texas - Austin $10,096 $10,490 $11,268 $14,435 $16,310 $20,364 $24,544 $27,760 $30,006
Rutgers
University of
New Jersey $15,829 $16,866 $17,715 $18,248 $19,753 $20,356 $22,059 $22,518 $24,044
From that, we can calculate the growth rate via ((new tuition price – old tuition)
Since it is clear that there is positive correlation between time and tuition price, the
best approach is graphing a time trend analysis rather than fitting the data to its
best-fit distribution. Inserting sparklines at the end of each row helps to show at a
glance the stability or instability of the growth rate over the period and better give
us an idea of what to expect when graphing the university tuitions (i.e. is the growth
19
rate steadily increasing, decreasing, or irregular?). To proceed with the graph,
tuition prices will be the Y-axis and the year will be the X-axis. Once all the data has
been graphed, each data series can be selected and the trendline that yields the
forward (to ensure that the projected data seems to fall in line with the original
data). For each data set, or university, the trendline type, equation (rounded), and
R2 is shown on the graph (Figure 3). As mentioned before, the coefficients need to
the graph, the straight-marked scatter plots shows our actual university prices over
8 years with each university identifiable in the legend, below which shows the
specific trendline best suited for each data series. There were two trendlines that
had the highest coefficient of determination, yet when taking a look at their future
Berkeley and for the University of Texas at Austin, both had a best fit of a third
degree polynomial function. However, the two-period forecast on the graph showed
U.C. Berkeley to quickly overtake the Ivy-League schools in terms of tuition price,
while U.T. Austin’s future tuition prices began to trend down significantly. Since
both of these seemed questionable, the next best fit of a logarithmic trendline was
chosen. Finally, the top right corner of the graph shows the prediction equations in
20
Figure 3: Tuition and Time-Trend Analysis
Because we’ve graphed 8 years of data for 10 different universities, as well as all the
trendlines and corresponding equations, the chart can seem overwhelming. It’s
easier to see the equations and their fit are shown in Table 8.
University of y = 3,792,826.2785ln(x) -
California-Berkeley 28,816,827.4477 Logarithmic 0.96507
21
Coefficient of
University Formula Type Determination
y = -60.2576x2 +
Pennsylvania State 243,009.7606x -
University 244,977,213.0485 Polynomial 0.99589
y = 3,431,268.8565ln(x) -
New York University 26,059,109.5563 Logarithmic 0.99811
Now, we can copy the formulas into the cells at the end our observed tuition
prices where X will be the year for the predicted tuition price and Y will be said
tuition. We have data up to 2013 for our federal subsidized and unsubsidized loan
rates and predicted the future loan rates out to 2017, but only have data up to 2010
for the tuition prices. This was the only 8-year period where all universities tuition
prices were given without some data missing for certain years. Ideally, the longer
the historical data the more accurate we can assume our prediction equation to be.
The tuition prices must be predicted out to year 2017 to fall in line with our
predicted interest rates, so our predicted values are nearly as many as our observed
values. It should be noted that the longer the prediction values compared to the
observed values, the higher the probability of marginal error A. The magnitude of
error can be measured by the mean absolute percentage error (MAPE), by using the
equations to predict back through the years that we have actual tuition prices and
then comparing the two. MAPE is the sum of the absolute value of the forecasted
data subtracted from the actual data over N, or the number of data used, multiplied
by 100 will yield the average percent our predictions are off. The equation is:
22
𝐹𝑜𝑟𝑒𝑐𝑎𝑠𝑡𝑒𝑑 𝐷𝑎𝑡𝑎 − 𝐴𝑐𝑡𝑢𝑎𝑙 𝐷𝑎𝑡𝑎
∑| |
𝐴= 𝐴𝑐𝑡𝑢𝑎𝑙 𝐷𝑎𝑡𝑎 × 100
𝑛
Table 9 shows the predicted values and can be compared to table 7 shown
University
2002 2003 2004 2005 2006 2007 2008 2009 2010
Princeton
University $25,945.09 $27,399.27 $28,770.88 $30,059.91 $31,266.36 $32,390.24 $33,431.55 $34,390.27 $35,266.42
Harvard
University $25,848.23 $27,552.99 $29,175.97 $30,717.17 $32,176.60 $33,554.24 $34,850.11 $36,064.19 $37,196.50
Dartmouth
College $26,175.53 $27,692.23 $29,208.93 $30,725.63 $32,242.33 $33,759.03 $35,275.73 $36,792.43 $38,309.13
Columbia
University $26,645.84 $28,129.90 $29,696.61 $31,350.58 $33,096.68 $34,940.02 $36,886.03 $38,940.42 $41,109.24
Yale
University $25,924.67 $27,284.34 $28,644.01 $30,003.67 $31,363.34 $32,723.01 $34,082.67 $35,442.34 $36,802.01
University
of
California-
Berkeley $15,866.07 $17,760.12 $19,653.22 $21,545.37 $23,436.59 $25,326.86 $27,216.18 $29,104.57 $30,992.02
Pennsylvan
ia State
University $15,723.02 $17,401.19 $18,958.84 $20,395.98 $21,712.61 $22,908.72 $23,984.31 $24,939.39 $25,773.95
New York
University $25,059.89 $26,773.38 $28,486.02 $30,197.80 $31,908.73 $33,618.81 $35,328.03 $37,036.41 $38,743.93
University
of Texas -
Austin $7,429.61 $10,166.31 $12,901.64 $15,635.60 $18,368.21 $21,099.45 $23,829.33 $26,557.85 $29,285.01
Rutgers
University
of New
Jersey $17,954.99 $18,903.90 $19,902.95 $20,954.81 $22,062.25 $23,228.23 $24,455.82 $25,748.29 $27,109.07
Now, when executing the formula mentioned previously, a value of 2.95 is yielded.
This tells us that our prediction equations compared to our actual observed data is
on average only 2.95% inaccurate. Using the tuition growth rates from Figure 2, the
mean average growth rate/year is 6.49%. Comparing this to our first model
university, it can be seen that by the fourth year the tuition is on average 25.96% off.
Therefore, the accuracy of the second model is greater than that of the first.
23
After we have plugged in the trendline equations, the following predicted
tuition prices are computed, of which we’ll be using 2014-2017 in calculating total
In moving along with the model, now that we have our predicted tuition
prices as well as predicted interest rate values, we can begin to calculate the total
costs as before in the first model. The total degree cost is calculated by summing the
predicted tuition prices from 2014-2017 for each university. The total loan amount
is slightly trickier to calculate. Since the subsidized loans do not accrue interest
while in school, you simply add up the percentage of the tuition allowed to be taken
out in subsidized loans for all four years attending college. As previously stated,
subsidized loans are not to exceed 63.64% of the tuition for the first year (2014),
69.23% for the second (2015), and 73.33% for the third and fourth year (2016 and
2017) of attendance. The remaining tuition (36.36% for 2014, 30.77% for 2015,
and 26.67% for 2016 and 2017) must be taken out in unsubsidized loans, which do
24
unsubsidized federal interest rates are used to find the future value of each loan.
The 2014 loan will be gathering interest for all for years of college attendance so the
calculation will be [(2014 Tuition * .3636)*(1.031^3.917)]. The 2015 loan will only
be gathering interest from the second year (date of disbursement) through 2017 so
same principle is applied to the last two loans with the calculations being [(2016
total loan amount, then, is figured by adding the future value of all unsubsidized
loans and the subsidized loans. Then, the total interest accrued during attendance is
simply computed by subtracting the degree cost from the total loan amount. Table
Table 11: Cost, Note Payable, and Interest Accrued During College
Total Interest
Total Degree Cost Total Note Payable Accrued During
University Upon Repayment College
Princeton University $154,230.42 $157,839.41 $3,608.99
Harvard University $167,645.06 $171,553.10 $3,908.04
Dartmouth College $186,603.93 $190,901.99 $4,298.06
Columbia University $221,952.31 $226,986.97 $5,034.66
Yale University $177,120.69 $181,205.55 $4,084.86
University of
California-Berkeley $165,422.57 $169,197.63 $3,775.05
Pennsylvania State
University $112,538.08 $115,182.00 $2,643.92
New York University $192,478.50 $196,902.84 $4,424.34
University of Texas -
Austin $177,037.43 $181,031.44 $3,994.01
Rutgers University of
New Jersey $144,181.02 $147,457.84 $3,276.82
25
Instead of using the average starting salary of each university, we’ll use the
In comparing the overall average salary per university with that of the student’s
intended degree salary, we can calculate the percentage change via ((Starting Salary
earlier, it was thought that there would most likely be a difference in a graduate’s
salary based upon their major. The salaries of lower paying majors taken into
account for each university’s overall average starting salary was assumed to likely
have “dragged down” the average salary when compared to a degree requiring more
technical skills. As can be seen from above, this is the case in all but one university.
Now, with all of the information thus far, we can complete the model and
calculate the monthly loan payment for the graduate (still assumed to be the
standard 10 year repayment plan), the total amount paid upon completion of
repayment, the total interest paid, the percent of the graduate’s monthly salary
26
spend on repayment, and their net disposable income per month as well as per
annum.
use variable rates, meaning the interest is subject to change every year. In order to
calculate the payments for the unsubsidized loans using a variable rate strategy, we
would have to use our trendline equation to predict the unsubsidized interest rates
10 years further into the future, which is nearly double the actual observed data and
therefore subject to much more error. From there we would then need to calculate
the payment for each year the interest rate changes and add them up (along with
the fixed interest subsidized loans) for the graduate’s expected monthly loan
payment. Since this would highly decrease the accuracy of the model, we will
assume the four predicted unsubsidized interest rates to be fixed from the date of
disbursement, much like the subsidized loans. Thus, to calculate the monthly loan
payment in Excel we use the payment function, which requires the interest rate,
number of payment periods, and the present value of the loan. The interest rate will
be the annual rate divided by 12 for the monthly interest. The number of payment
periods is the same as in the first model (120 months, or 10 years) and the present
value (which must be input as a negative) is the amount of the loan at each interest
rate upon graduation. So, we must sum the payment calculations of each loan that
has a different interest rate in order to get the final monthly loan payment the
27
student will incur for the 10 years following graduation (or until full loan
(2014Tuition*63.64%+2015Tuition*69.23%+2016Tuition*73.33%+2017Tuition*7
3.33%) for all of the subsidized loans taken out during the four year period at the
static interest rate of 3.86%. To this must be added the payments for each
(2014Tuition*36.36%)*(1.031^3.917)) + PMT((2.91%/12),120,-
(2015Tuition*30.77%)*(1.0291^2.917)+ PMT((2.80%/12),120,-
(2016Tuition*26.67%)*(1.028^1.917)+ PMT((2.75%/12),120,-
loan payment. From there, to find the total amount paid after 10 years, you simply
multiply the monthly loan payment by 120 months. To find the total interest paid,
you subtract the total amount paid after 10 years from the total degree cost.
After this, the graduate’s starting salary can now be taken into account to see
the feasibility of loan repayment. To calculate the percent of monthly salary spent
on loans, first divide the average annual starting salary by 12 to find the monthly
salary, then divide the monthly loan payment by the monthly salary and multiply by
100 to get a percentage. For the net disposable income the graduate can expect per
month during repayment, simply subtract the monthly loan payment from the
monthly salary, and for per annum disposable income merely multiply the latter by
12. After completing all of the above calculations, we will have generated a
28
reasonable overview for expected expenses and disposable salary until full loan
repayment.
29
Analysis and Limitations of Model 2
As the table above shows, this model shows a more favorable outlook on the
cost of college compared to salary after graduation. Whereas the first model had
going towards their loan repayment, this model shows that a engineering graduate
from any of the above colleges can expect to pay anywhere from about a quarter to
less than half of their salary on their student loans. Should the prospective student
wish to consider a different university from the ones listed above, it is a simple
matter of finding the historical tuition rate for any 8 year period (preferably nearer
to their expected year of entry to the college), as well as the starting salary for their
intended major from that university. After inputting the historical tuition, the graph
of tuition prices will automatically update, however the trendline will not. The user
must manually pick the trendline with the highest coefficient of determination and
then copy the formula given on the graph into the cells they wish to predict future
tuition rates. Another alternative is to simply enter directly into the forecasting
cells reasonable forecasted tuition. From there, all the information will update and
There are some limitations to this model, as was the case in the first. As
mentioned earlier, the relatively small amount of historical data used for the time-
trend analysis, both for tuition prices and interest rates, can pose a problem the
further ahead the user attempts to forecast. A trendline could be more heavily
relied upon when there is a larger quantity of historical data to fit it to. The interest
rates, since there was again too little historical data on the government website,
30
yielded a trendline with the lowest coefficient of determination of all the trendlines
used in the model. Because of this, it was safer to forecast only 4 years in advance
and treat the unsubsidized federal interest rates as fixed-rate loans in calculating
the monthly loan payments for the 10 years following graduation, rather than
compute the monthly loan payments for the unsubsidized loans as a variable-rate
31
Conclusions
In conclusion, we’ve improved this model from the first by increasing the
complexity of it. The second model is more realistic than the first for a few reasons.
Firstly, by not assuming static variables, but by taking historical data to forecast a
likely future rate in interest and in tuition price. In addition, instead of taking the
university’s overall average starting salary, we chose to use the starting salary for
the fictitious student’s planned major (engineering). Also, instead of assuming the
student will be responsible for the full financial burden of college, the assumption
was changed to a more likely scenario in which the student’s family contributes
contribution (EFC) was paying for room and board, books, meal plans, and other
miscellaneous fees, while the student was responsible for tuition payment. We also
split the loan amounts based on real percentages given by FinAid between Federal
Subsidized and Federal Unsubsidized, instead of assuming as in the first model that
the student would only have Federal unsubsidized loans, drastically reducing the
amount of interest accrued during college attendance. From all this, we were able
to forecast a financial breakdown for the student that was much less overwhelming
than in the first model as well as much more realistic. This model is by no means
perfect, and there is plenty of room for further improvement. For instance, taking
into consideration the economy could prove to be a very useful factor that could be
included in this model with regards to inflation rates in predicting more accurate
interest rates as well as future expected salaries. Including income tax rates would
help by generating a net income versus a gross income for the user. Additionally,
32
perhaps the prospective student has a small sum of money and would like to see
whether they should simply take out less loans, or if investing in the stock market
may yield them a higher return rate over their loan rates, thereby enabling them to
pay off more of their loans by investing rather than simply putting their sum
towards tuition to begin with. Merely choosing a couple of stocks and their
historical annual returns over a period would allow you to calculate the mean
annual return and standard deviation for each. Then, by calculating the return and
standard deviation of investing 0% in the first stock and 100% in the second, all the
way through to 100% in the first and 0% in the second, a graph of a parabola can be
produced of which the minimum point represents the highest possible return with
the lowest possible risk (standard deviation) that corresponds to the weight
invested in each stock (i.e. 36% Stock 1, 64% Stock 2) as seen in Figure 4.
0.12
0.1
0.08
return
0.06
0.04
0.02
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Risk (Standard Deviation)
There are many more ways in which to improve upon the model, but the
common theme is they all require more data and fewer assumptions. Therefore, it
33
can be concluded that increasing model complexity by taking into account more
variables, and therefore a larger quantity of data, will yield a more accurate result in
future prognostications.
34
References
"2014 Poverty Guidelines." U.S. Department of Health & Human Services. ASPE, 22
"CNN Money." How Much Will That College Really Cost? N.d.. Web. 19 Mar 2014.
<http://cgi.money.cnn.com/tools/collegecost/collegecost.html>.
"Cost of Tuition for American Colleges and Universities." CollegeCalc. N.d.. Web. 19
Gage, Caroline, and Janet Lorin. "Student Loans, the Next Big Threat to the U.S.
2014. <http://www.businessweek.com/articles/2014-01-16/student-loans-
the-next-big-threat-to-the-u-dot-s-dot-economy>.
Jay, John. "The World According to Quants: Enter Alpha Generation Platforms." Wall
Street and Technology. Wall Street & Technology, 14 July 2008. Web. 19
quants-enter-alph/209000157>.
Pay Scale, . "2012-2013 PayScale College Salary Report." N.d.. Web. 19 Mar 2014.
<http://www.payscale.com/college-salary-report-2013/full-list-of-schools>.
<http://www.finaid.org/loans/repayment.phtml>.
calculated and what fees are associated with your federal student loan. N.d.
35
Yates, Tristian. "Quants: The Rocket Scientists Of Wall Street,” 10 May 2013. Web. 18
Mar 2014.
<http://www.investopedia.com/articles/financialcareers/08/quants-
quantitative-analyst.asp>.
36
Appendix
37
Total
Average Total Note
Interest
Annual Total Payable
University Accrued
Cost of Degree Cost Upon
During
Attendance Repayment
College
University of
Texas -
Austin $25,587.00 $102,348.00 $112,519.79 $10,171.79
Rutgers
University of
New Jersey $29,512.00 $118,048.00 $129,780.12 $11,732.12
38
Table 4: Effect on Salary
Net
Average Monthly % Monthly Disposable
Monthly Loan Salary To Income
University Salary Payment Loan Month Per Year
Princeton
University $4,858.33 $2,548.94 52.47% $2,309.39 $27,712.73
Harvard
University $4,225.00 $2,664.47 63.06% $1,560.53 $18,726.35
Dartmouth
College $4,508.33 $2,832.28 62.82% $1,676.05 $20,112.61
Columbia
University $4,558.33 $2,872.75 63.02% $1,685.58 $20,226.95
Yale University $4,075.00 $2,712.81 66.57% $1,362.19 $16,346.22
University of
California-
Berkeley $4,341.67 $1,436.58 33.09% $2,905.09 $34,861.09
Pennsylvania
State University $4,050.00 $1,394.03 34.42% $2,655.97 $31,871.69
New York
University $4,058.33 $2,888.41 71.17% $1,169.92 $14,039.06
University of
Texas - Austin $4,066.67 $1,131.74 27.83% $2,934.93 $35,219.16
Rutgers
University of
New Jersey $4,041.67 $1,305.34 32.30% $2,736.32 $32,835.89
39
Figure 1: Federal Unsubsidized Loan Rates and Time-trend Analysis
40
Figure 2: University Tuition Growth Rate From 2003-2010
41
Table 8: Trendline Equation, Type, and R2
Coefficient of
University Formula Type Determination
y = -41.2879x2 +
Princeton University 166,812.1364x -
168,449,968.5758 Polynomial 0.99084
y = -40.8896x2 +
Harvard University 165,467.6502x -
167,354,665.5030 Polynomial 0.99867
y = 1,516.7x -
Dartmouth College
3,010,257.8667 Linear 0.99616
Columbia University y = 2*10-43e0.0542x Exponential 0.99968
Yale University y = 1,359.6667x - 2,696,128 Linear 0.99495
University of y = 3,792,826.2785ln(x) -
California-Berkeley 28,816,827.4477 Logarithmic 0.96507
y = -60.2576x2 +
Pennsylvania State
243,009.7606x -
University
244,977,213.0485 Polynomial 0.99589
University
2002 2003 2004 2005 2006 2007 2008 2009 2010
Princeton
University $25,945.09 $27,399.27 $28,770.88 $30,059.91 $31,266.36 $32,390.24 $33,431.55 $34,390.27 $35,266.42
Harvard
University $25,848.23 $27,552.99 $29,175.97 $30,717.17 $32,176.60 $33,554.24 $34,850.11 $36,064.19 $37,196.50
Dartmouth
College $26,175.53 $27,692.23 $29,208.93 $30,725.63 $32,242.33 $33,759.03 $35,275.73 $36,792.43 $38,309.13
Columbia
University $26,645.84 $28,129.90 $29,696.61 $31,350.58 $33,096.68 $34,940.02 $36,886.03 $38,940.42 $41,109.24
Yale
University $25,924.67 $27,284.34 $28,644.01 $30,003.67 $31,363.34 $32,723.01 $34,082.67 $35,442.34 $36,802.01
University
of
California-
Berkeley $15,866.07 $17,760.12 $19,653.22 $21,545.37 $23,436.59 $25,326.86 $27,216.18 $29,104.57 $30,992.02
Pennsylvan
ia State
University $15,723.02 $17,401.19 $18,958.84 $20,395.98 $21,712.61 $22,908.72 $23,984.31 $24,939.39 $25,773.95
New York
University $25,059.89 $26,773.38 $28,486.02 $30,197.80 $31,908.73 $33,618.81 $35,328.03 $37,036.41 $38,743.93
University
of Texas -
Austin $7,429.61 $10,166.31 $12,901.64 $15,635.60 $18,368.21 $21,099.45 $23,829.33 $26,557.85 $29,285.01
Rutgers
University
of New
Jersey $17,954.99 $18,903.90 $19,902.95 $20,954.81 $22,062.25 $23,228.23 $24,455.82 $25,748.29 $27,109.07
42
Table 10: Predicted Tuition
Tuition
University 2011 2012 2013 2014 2015 2016 2017
Princeton University $36,060.00 $36,771.00 $37,399.42 $37,945.27 $38,408.55 $38,789.24 $39,087.36
Harvard University $38,247.02 $39,215.77 $40,102.74 $40,907.93 $41,631.34 $42,272.97 $42,832.82
Dartmouth College $39,825.83 $41,342.53 $42,859.23 $44,375.93 $45,892.63 $47,409.33 $48,926.03
Columbia University $43,398.85 $45,815.98 $48,367.73 $51,061.61 $53,905.52 $56,907.83 $60,077.35
Yale University $38,161.67 $39,521.34 $40,881.01 $42,240.67 $43,600.34 $44,960.01 $46,319.67
University of California-Berkeley $32,878.53 $34,764.10 $36,648.73 $38,532.43 $40,415.20 $42,297.02 $44,177.92
Pennsylvania State University $26,488.00 $27,081.53 $27,554.55 $27,907.05 $28,139.04 $28,250.52 $28,241.47
New York University $40,450.61 $42,156.43 $43,861.41 $45,565.54 $47,268.83 $48,971.27 $50,672.86
University of Texas - Austin $32,010.82 $34,735.27 $37,458.37 $40,180.12 $42,900.51 $45,619.55 $48,337.25
Rutgers University of New Jersey $28,541.76 $30,050.17 $31,638.29 $33,310.35 $35,070.78 $36,924.24 $38,875.65
Table 11: Cost, Note Payable, and Interest Accrued During College
Total Interest
Total Degree Cost Total Note Payable Accrued During
University Upon Repayment College
Princeton University $154,230.42 $157,839.41 $3,608.99
Harvard University $167,645.06 $171,553.10 $3,908.04
Dartmouth College $186,603.93 $190,901.99 $4,298.06
Columbia University $221,952.31 $226,986.97 $5,034.66
Yale University $177,120.69 $181,205.55 $4,084.86
University of
California-Berkeley $165,422.57 $169,197.63 $3,775.05
Pennsylvania State
University $112,538.08 $115,182.00 $2,643.92
New York University $192,478.50 $196,902.84 $4,424.34
University of Texas -
Austin $177,037.43 $181,031.44 $3,994.01
Rutgers University of
New Jersey $144,181.02 $147,457.84 $3,276.82
43
Table 12: Change in Average Salary and BS/BA Salary
Overall Average
Average Starting
University
Starting Salary with Percent
Salary BS/BA Increase/Decrease
Princeton University $58,300 $67,138 15.16%
Harvard University $50,700 $88,437 74.43%
Dartmouth College $54,100 $73,698 36.23%
Columbia University $54,700 $56,595 3.46%
Yale University $48,900 $68,747 40.59%
University of California-Berkeley $52,100 $57,109 9.61%
Pennsylvania State University $48,600 $52,236 7.48%
New York University $48,700 $50,111 2.90%
University of Texas - Austin $48,800 $60,429 23.83%
Rutgers University of New Jersey $48,500 $46,333 -4.47%
44
Figure 4: 2-Stock Optimization
0.12
0.1
0.08
return
0.06
0.04
0.02
0
0 0.05 0.1 0.15 0.2 0.25 0.3
Risk (Standard Deviation)
45