OK. I'm brand new to this site so "Hello All"! Well I've been wrestling with a difficult problem for the last week and I would appreciate any help you can give me.
I know there are many formulas out there to calculate APR but I've tested many formulas and they do not handle Odd-Days properly for closed-end (consumer loans). The government has attempted to give us mere mortals some help with this by publishing an Appendix J to their truth-in-lending act.
It can be found here: https://www.fdic.gov/regulations/laws/rules/6500-3550.html
If you're brave (!!), you can see the formulas they provide which will solve for the APR, including the Odd-Days of the loan. Odd-Days are the days at the beginning of the loan that isn't really covered by a regular period payment but interest is still being charged. For example, you take a loan for $1,000.00 on 01/20/2012 and your first payment is 03/01/2012. You have 10 odd-days from 01/20/2012 to 01/30/2012. All months are 30 days for their calcs.
What I'm hoping for is someone with a significant background in Calculus who can interpret the the formulas you'll find about half way down Appendix J. And interpret the Actuarial method they're using to solve these formulas. I understand the iterative process. I first tried to solve this using the Newton-Raphson method but my formula for the APR did not account for the Odd-days. It worked great in the unlikely trivial case where there are no odd days, but struggled with odd-days.
I know that reading this document is very difficult! I've made some headway but there are certain things I just can't figure out how they're doing. They seem to introduce a few things as if by magic.
Anyways thanks ahead of time for helping! :)
Alright, you weren't kidding about the document being a bit hard to read. The solution is actually not that bad though, depending on implementation. I failed repeatedly trying to use their various simplified forumulae and eventually got it using the general formula up top(8). Technically this is a simplification. The actual general formula would take arrays of length period for the other arguments and use their indexes in the loop. You use this method to get A' and A'' for the iteration step. Odd days are handled by (1.0 + fractions*rate), which appears as 1 + f i in the document. Rate is the rate per period, not overall apr.
public double generalEquation(int period, double payment, double initialPeriods, double fractions, double rate)
{
double retval = 0;
for (int x = 0; x < period; x++)
retval += payment / ((1.0 + fractions*rate)*Math.pow(1+rate,initialPeriods + x));
return retval;
}
Iteration behaves just as the document says in its example(9).
/**
*
* @param amount The initial amount A
* @param payment The periodic payment P
* @param payments The total number of payments n
* @param ppy The number of payment periods per year
* @param APRGuess The guess to start estimating from, 10% is 0.1, not 0.001
* @param partial Odd days, as a fraction of a pay period. 10 days of a month is 0.33333...
* @param full Full pay periods before the first payment. Usually 1.
* @return The calculated APR
*/
public double findAPRGEQ(double amount, double payment, int payments, double ppy, double APRGuess, double partial, double full)
{
double result = APRGuess;
double tempguess = APRGuess;
do
{
result = tempguess;
//Step 1
double i = tempguess/(100*ppy);
double A1 = generalEquation(payments, payment, full, partial, i);
//Step 2
double i2 = (tempguess + 0.1)/(100*ppy);
double A2 = generalEquation(payments, payment, full, partial, i2);
//Step 3
tempguess = tempguess + 0.1*(amount - A1)/(A2 - A1);
System.out.println(tempguess);
} while (Math.abs(result*10000 - tempguess*10000) > 1);
return result;
}
Note that as a general rule it is BAD to use double for monetary calculations as I have done here, but I'm writing a SO example, not production code. Also, it's java instead of .net, but it should help you with the algorithm.
Tho this is an old thread, I'd like to help others avoid wasting time on this - translating the code to PHP (or even javascript), gives wildly inaccurate results, causing me to wonder if it really worked in Java -
<?php
function generalEquation($period, $payment, $initialPeriods, $fractions, $rate){
$retval = 0;
for ($x = 0; $x < $period; $x++)
$retval += $payment / ((1.0 + $fractions*$rate)*pow(1+$rate,$initialPeriods + $x));
return $retval;
}
/**
*
* @param amount The initial amount A
* @param payment The periodic payment P
* @param payments The total number of payments n
* @param ppy The number of payment periods per year
* @param APRGuess The guess to start estimating from, 10% is 0.1, not 0.001
* @param partial Odd days, as a fraction of a pay period. 10 days of a month is 0.33333...
* @param full Full pay periods before the first payment. Usually 1.
* @return The calculated APR
*/
function findAPR($amount, $payment, $payments, $ppy, $APRGuess, $partial, $full)
{
$result = $APRGuess;
$tempguess = $APRGuess;
do
{
$result = $tempguess;
//Step 1
$i = $tempguess/(100*$ppy);
$A1 = generalEquation($payments, $payment, $full, $partial, $i);
//Step 2
$i2 = ($tempguess + 0.1)/(100*$ppy);
$A2 = generalEquation($payments, $payment, $full, $partial, $i2);
//Step 3
$tempguess = $tempguess + 0.1*($amount - $A1)/($A2 - $A1);
} while (abs($result*10000 - $tempguess*10000) > 1);
return $result;
}
// these figures should calculate to 12.5 apr (see below)..
$apr = findAPR(10000,389.84,(30*389.84),12,.11,0,1);
echo "APR: $apr" . "%";
?>
APR: 12.5000%
Total Financial Charges: $1,695.32
Amount Financed: $10,000.00
Total Payments: $11,695.32
Total Loan: $10,000.00
Monthly Payment: $389.84
Total Interest: $1,695.32
I got me a Python (3.4) translation here. And since my application takes dates as inputs, not full and partial payment periods, I threw in a way to calculate those. I referenced a document by one of the guys that wrote the OCC's APRWIN, and I'd advise others to read it if you need to re-translate this.
My tests come straight from the Reg Z examples. I haven't done further testing with APRWIN yet. An edge case I don't have to deal with (so haven't coded for) is when you only have 2 installments and the first is an irregular period. Check the document above if that's a potential use case for your app. I also haven't fully tested most of the payment schedules because my app only needs monthly and quarterly. The rest are just there to use Reg Z's examples.
# loan_amt: initial amount of A
# payment_amt: periodic payment P
# num_of_pay: total number of payment P
# ppy: number of payment periods per year
# apr_guess: guess to start estimating from. Default = .05, or 5%
# odd_days: odd days, meaning the fraction of a pay period for the first
# installment. If the pay period is monthly & the first installment is
# due after 45 days, the odd_days are 15/30.
# full: full pay periods before the first payment. Usually 1
# advance: date the finance contract is supposed to be funded
# first_payment_due: first due date on the finance contract
import datetime
from dateutil.relativedelta import relativedelta
def generalEquation(period, payment_amt, full, odd_days, rate):
retval = 0
for x in range(period):
retval += payment_amt / ((1.0 + odd_days * rate) * ((1 + rate) ** (
x + full)))
return retval
def _dt_to_int(dt):
"""A convenience function to change datetime objects into a day count,
represented by an integer"""
date_to_int = datetime.timedelta(days=1)
_int = int(dt / date_to_int)
return _int
def dayVarConversions(advance, first_payment_due, ppy):
"""Takes two datetime.date objects plus the ppy and returns the remainder
of a pay period for the first installment of an irregular first payment
period (odd_days) and the number of full pay periods before the first
installment (full)."""
if isinstance(advance, datetime.date) and isinstance(first_payment_due,
datetime.date):
advance_to_first = -relativedelta(advance, first_payment_due)
# returns a relativedelta object.
## Appendix J requires calculating odd_days by counting BACKWARDS
## from the later date, first subtracting full unit-periods, then
## taking the remainder as odd_days. relativedelta lets you
## calculate this easily.
# advance_date = datetime.date(2015, 2, 27)
# first_pay_date = datetime.date(2015, 4, 1)
# incorrect = relativedelta(first_pay_date, advance_date)
# correct = -relativedelta(advance_date, first_pay_date)
# print("See the difference between ", correct, " and ", incorrect, "?")
if ppy == 12:
# If the payment schedule is monthly
full = advance_to_first.months + (advance_to_first.years * 12)
odd_days = advance_to_first.days / 30
if odd_days == 1:
odd_days = 0
full += 1
# Appendix J (b)(5)(ii) requires the use of 30 in the
# denominator even if a month has 31 days, so Jan 1 to Jan 31
# counts as a full month without any odd days.
return full, odd_days
elif ppy == 4:
# If the payment schedule is quarterly
full = (advance_to_first.months // 3) + (advance_to_first.years * 4)
odd_days = ((advance_to_first.months % 3) * 30 + advance_to_first. \
days) / 90
if odd_days == 1:
odd_days = 0
full += 1
# Same as above. Sometimes odd_days would be 90/91, but not under
# Reg Z.
return full, odd_days
elif ppy == 2:
# Semiannual payments
full = (advance_to_first.months // 6) + (advance_to_first.years * 2)
odd_days = ((advance_to_first.months % 6) * 30 + advance_to_first. \
days) / 180
if odd_days == 1:
odd_days = 0
full += 1
return full, odd_days
elif ppy == 24:
# Semimonthly payments
full = (advance_to_first.months * 2) + (advance_to_first.years * \
24) + (advance_to_first.days // 15)
odd_days = ((advance_to_first.days % 15) / 15)
if odd_days == 1:
odd_days = 0
full += 1
return full, odd_days
elif ppy == 52:
# If the payment schedule is weekly, then things get real
convert_to_days = first_payment_due - advance
# Making a timedelta object
days_per_week = datetime.timedelta(days=7)
# A timedelta object equal to 1 week
if advance_to_first.years == 0:
full, odd_days = divmod(convert_to_days, days_per_week)
# Divide, save the remainder
odd_days = _dt_to_int(odd_days) / 7
# Convert odd_days from a timedelta object to an int
return full, odd_days
elif advance_to_first.years != 0 and advance_to_first.months == 0 \
and advance_to_first.days == 0:
# An exact year is an edge case. By convention, we consider
# this 52 weeks, not 52 weeks & 1 day (2 if a leap year)
full = 52 * advance_to_first.years
odd_days = 0
return full, odd_days
else:
# For >1 year, there need to be exactly 52 weeks per year,
# meaning 364 day years. The 365th day is a freebie.
year_remainder = convert_to_days - datetime.timedelta(days=(
365 * advance_to_first.years))
full, odd_days = divmod(year_remainder, days_per_week)
full += 52 * advance_to_first.years
# Sum weeks from this year, weeks from past years
odd_days = _dt_to_int(odd_days) / 7
# Convert odd_days from a timedelta object to an int
return full, odd_days
else:
print("What ppy was that?")
### Raise an error appropriate to your application
else:
print("'advance' and 'first_payment_due' should both be datetime.date objects")
def regulationZ_APR(loan_amt, payment_amt, num_of_pay, ppy, advance,
first_payment_due, apr_guess=.05):
"""Returns the calculated APR using Regulation Z/Truth In Lending Appendix
J's calculation method"""
result = apr_guess
tempguess = apr_guess + .1
full, odd_days = dayVarConversions(advance, first_payment_due, ppy)
while abs(result - tempguess) > .00001:
result = tempguess
# Step 1
rate = tempguess/(100 * ppy)
A1 = generalEquation(num_of_pay, payment_amt, full, odd_days, rate)
# Step 2
rate2 = (tempguess + 0.1)/(100 * ppy)
A2 = generalEquation(num_of_pay, payment_amt, full, odd_days, rate2)
# Step 3
tempguess = tempguess + 0.1 * (loan_amt - A1)/(A2 - A1)
return result
import unittest
class RegZTest(unittest.TestCase):
def test_regular_first_period(self):
testVar = round(regulationZ_APR(5000, 230, 24, 12,
datetime.date(1978, 1, 10), datetime.date(1978, 2, 10)), 2)
self.assertEqual(testVar, 9.69)
def test_long_first_payment(self):
testVar = round(regulationZ_APR(6000, 200, 36, 12,
datetime.date(1978, 2, 10), datetime.date(1978, 4, 1)), 2)
self.assertEqual(testVar, 11.82)
def test_semimonthly_payment_short_first_period(self):
testVar = round(regulationZ_APR(5000, 219.17, 24, 24,
datetime.date(1978, 2, 23), datetime.date(1978, 3, 1)), 2)
self.assertEqual(testVar, 10.34)
def test_semimonthly_payment_short_first_period2(self):
testVar = round(regulationZ_APR(5000, 219.17, 24, 24,
datetime.date(1978, 2, 23), datetime.date(1978, 3, 1), apr_guess=
10.34), 2)
self.assertEqual(testVar, 10.34)
def test_quarterly_payment_long_first_period(self):
testVar = round(regulationZ_APR(10000, 385, 40, 4,
datetime.date(1978, 5, 23), datetime.date(1978, 10, 1), apr_guess=
.35), 2)
self.assertEqual(testVar, 8.97)
def test_weekly_payment_long_first_period(self):
testVar = round(regulationZ_APR(500, 17.6, 30, 52,
datetime.date(1978, 3, 20), datetime.date(1978, 4, 21), apr_guess=
.1), 2)
self.assertEqual(testVar, 14.96)
class dayVarConversionsTest(unittest.TestCase):
def test_regular_month(self):
full, odd_days = dayVarConversions(datetime.date(1978, 1, 10), datetime.date(
1978, 2, 10), 12)
self.assertEqual(full, 1)
self.assertEqual(odd_days, 0)
def test_long_month(self):
full, odd_days = dayVarConversions(datetime.date(1978, 2, 10), datetime.date(
1978, 4, 1), 12)
self.assertEqual(full, 1)
self.assertEqual(odd_days, 19/30)
def test_semimonthly_short(self):
full, odd_days = dayVarConversions(datetime.date(1978, 2, 23), datetime.date(
1978, 3, 1), 24)
self.assertEqual(full, 0)
self.assertEqual(odd_days, 6/15)
def test_quarterly_long(self):
full, odd_days = dayVarConversions(datetime.date(1978, 5, 23), datetime.date(
1978, 10, 1), 4)
self.assertEqual(full, 1)
self.assertEqual(odd_days, 39/90)
def test_weekly_long(self):
full, odd_days = dayVarConversions(datetime.date(1978, 3, 20), datetime.date(
1978, 4, 21), 52)
self.assertEqual(full, 4)
self.assertEqual(odd_days, 4/7)
