DrawdownPeak Function Has Issues

[rjvelasquezm]

Noticed an issue in the DrawdownPeak private function when trying to calculate a Ulcer Index myself using the Drawdowns function.

Seems to me DrawdownPeak may be an old implementation of Drawdowns that never got swapped in some older functions.

Issue is returns are divided by 100 at one point, compounded and then multiplied back by 100. So it seems to assumes percent returns are passed in as integers instead of decimals. It seems to me in the rest of the package decimal returns are always expected so seems to fall out of place.

The fix is simple, just remove the divide by 100 and multiply by 100. That makes drawdown results match those returned by the Drawdowns function.

                                    val = val * (1 + R[j]/100)
                          
                                    drawdownpeak[i] = (val - 1) * 100

[femtotrader]

Found this issue also.

Description

The DrawdownPeak function in PerformanceAnalytics divides returns by 100 internally (line: val = val * (1 + R[j]/100)), which assumes returns are in percentage format (e.g., 1.5 for 1.5%). However, the standard convention in R (and the managers dataset included with PerformanceAnalytics) uses decimal returns (e.g., 0.015 for 1.5%).

This causes PainIndex (which uses DrawdownPeak) to compute incorrect values when given decimal returns.

Reproducible Example

library(PerformanceAnalytics)

# Load the managers dataset (decimal returns)
data(managers)
returns <- as.numeric(na.omit(managers[1:120, "HAM1"]))

# Check return format - they are decimals
cat("Sample returns:", head(returns), "\n")
# Output: 0.0074 0.0193 0.0155 -0.0091 0.0076 -0.0039

# R's PainIndex
returns_xts <- xts(returns, order.by = seq(as.Date("1996-01-31"), by = "month", length.out = length(returns)))
pain_r <- as.numeric(PainIndex(returns_xts))
cat("R PainIndex:", pain_r, "\n")
# Output: 0.0167311

# Correct PainIndex (manual calculation)
prices <- cumprod(1 + returns)
peak_prices <- cummax(prices)
drawdowns <- (prices - peak_prices) / peak_prices
pain_correct <- mean(abs(drawdowns))
cat("Correct PainIndex:", pain_correct, "\n")
# Output: 0.01715642

# Relative difference
cat("Relative error:", abs(pain_r - pain_correct) / pain_correct * 100, "%\n")
# Output: 2.54%

Root Cause

In DrawdownPeak, line 14:

val = val * (1 + R[j]/100)

This divides the return by 100, which is incorrect for decimal returns. For a return of 0.0074 (0.74%), this computes:

• Incorrect: 1 + 0.0074/100 = 1.000074
• Correct: 1 + 0.0074 = 1.0074

Impact

• PainIndex returns incorrect values (~2.5% error for typical data)
• PainRatio is also affected (uses PainIndex)
• Any analysis relying on these metrics will have systematic errors

Suggested Fix

Remove the /100 division in DrawdownPeak:

# Before (incorrect for decimal returns)
val = val * (1 + R[j]/100)

# After (correct for decimal returns)
val = val * (1 + R[j])

Alternatively, add a parameter to specify the return format, but this would be inconsistent with all other functions in the package that expect decimal returns.

Environment

• R version: 4.2.2
• PerformanceAnalytics version: 2.0.8

Related

This bug affects:

• DrawdownPeak()
• PainIndex()
• PainRatio()

[femtotrader]

Looks duplicate of #102