Calculate the Scorecard Points for a Class in an Independent Variable

points() calculates the number of scorecard points for a unique class in an independent variable using Weights of Evidence, helping to build a full "scorecard" of a number of points mapped to each class in each independent variable.

Usage

points(
  woe,
  estimate,
  intercept,
  num_vars,
  tgt_points,
  tgt_odds,
  pxo,
  rate = 2,
  round = 0
)

Arguments

woe: (Numeric) The Weight-of-Evidence value for a given class of the independent variable
estimate: (Numeric) The coefficient of the logistic regression model for the independent variable
intercept: (Numeric) The intercept value of the logistic regression model (where model is trained to predict the probability of "bad")
num_vars: (Integer) The number of independent variables in the logistic regression model
tgt_points: (Integer) The target number of points to be used in conjunction with the target odds; see the Details section of ?odds for more information
tgt_odds: (Numeric) The odds that the tgt_points should have; see the Details section of ?odds for more information
pxo: (Integer) The number of points to 'double' the odds; see the Details section of ?odds for more information
rate: (Numeric) The value to exponentially increase the odds by for the given number of points supplied in the pxo argument; see the Details section of ?odds for more information
round: (Integer) The number of digits to round the output score to; default is to round to the nearest integer

Value

A numeric value representing the number of scorecard points for the given class in the independent variable

Details

The tgt_points and tgt_odds arguments work together to build a "baseline" points/odds for your scorecard. For example, a tgt_points value of 300 and a tgt_odds value of 30 would be interpreted as a score of 300 points would have 30:1 odds of default. See Details section of ?odds for more information.

The intercept value must be from a glm model trained to predict probability of "bad". This means that the order of levels in the dependent variable must be c("good", "bad"). See ?binomial for more details about the order in which glm(family = "binomial") expects the levels of the dependent variable.

References

Siddiqi, Naeem (2017). Intelligent Credit Scoring: Building and Implementing Better Credit Risk Scorecards. 2nd ed., Wiley. pp. 240-242.

Examples

# Pre-process the data to create WoE features
df <- woe(
  data = loans |>
    dplyr::mutate(
      default_status = factor(loans$default_status, levels = c("good", "bad"))
    ),
  outcome = default_status,
  predictors = c(industry, housing_status),
  method = "replace",
  verbose = FALSE
)

# Fit the logistic regression model
fit <- glm(default_status ~  ., data = df, family = "binomial")

# Extract the model's parameter estimates & intercept
params <- fit$coefficients |>
  tibble::as_tibble(rownames = NA) |>
  tibble::rownames_to_column(var = "variable")

# Build the scorecard base
card <- woe(
  data = loans,
  outcome = default_status,
  predictors = c(industry, housing_status),
  method = "dict",
  verbose = FALSE
) |>
  dplyr::transmute(
    variable = paste0("woe_", variable),
    class = class,
    woe = woe
  ) |>
  dplyr::inner_join(params, by = "variable")

# Add the points
card |>
  dplyr::mutate(
    points = points(
      woe = woe,
      estimate = value,
      intercept = params$value[params$variable == "(Intercept)"],
      num_vars = length(params$variable[params$variable != "(Intercept)"]),
      tgt_points = 300L,
      tgt_odds = 30,
      pxo = 20L,
      rate = 2
    )
  )
#> # A tibble: 12 × 5
#>    variable           class            woe  value points
#>    <chr>              <chr>          <dbl>  <dbl>  <dbl>
#>  1 woe_industry       ""           -1.23   -0.991     78
#>  2 woe_industry       "beef"        0.231  -0.991    120
#>  3 woe_industry       "dairy"       0.0956 -0.991    116
#>  4 woe_industry       "fruit"       0.359  -0.991    123
#>  5 woe_industry       "grain"      -0.410  -0.991    101
#>  6 woe_industry       "greenhouse"  0.511  -0.991    128
#>  7 woe_industry       "nuts"        0.288  -0.991    121
#>  8 woe_industry       "pork"        0.606  -0.991    130
#>  9 woe_industry       "poultry"    -0.774  -0.991     91
#> 10 woe_industry       "sod"         0.154  -0.991    118
#> 11 woe_housing_status "own"        -0.194  -0.999    108
#> 12 woe_housing_status "rent"        0.430  -0.999    126

# Calculate points manually
points(
  woe = 1.23,
  estimate = -0.991,
  intercept = -0.846,
  num_vars = 2L,
  tgt_points = 300L,
  tgt_odds = 30,
  pxo = 20L,
  rate = 2
)
#> [1] 148