Downside risk

Standard deviation punishes a fund for going up as hard as for going down. No investor feels that way.

Downside risk is the probability of a loss on account of reducing asset prices in changing market conditions — the workbook's own footnote definition, and the idea behind the two measures that follow it in the chapter: maximum drawdown, which looks backwards at the worst fall that actually happened, and value at risk, which looks forwards at the worst fall expected at a given confidence level.

Maximum drawdown (MDD) is the peak-to-trough decline in the assets under management of an AIF during a specific reporting period, usually quoted as a percentage of the peak value achieved in that period.

It measures the extent of the greatest loss until a new peak is created — and the workbook is careful about what it does not do: it does not take into account how often the portfolio experiences large losses, and it does not indicate how long it will take an investor to recover. Investors would prefer a zero MDD, but the number is meaningless without the time period being considered and the performance of a benchmark over the same period.

Value at risk (VaR) is defined as the maximum amount of expected loss for a fund over a given time frame at a pre-defined confidence level. If the 95% one-month VaR of a fund is Rs 1 crore, there is a 95% probability that over the next month the fund will not lose more than Rs 1 crore. Its limitation is precise: while quantifying the maximum potential loss, VaR fails to indicate the size of the loss to be borne beyond that confidence level.

Downside risk also runs through the risk register itself. Leverage magnifies it; concentration makes a single name capable of producing it; and the workbook notes that certain risks — macro-economic, legal and regulatory, operational, country-specific and cyber security — are simply not measurable, whatever the metric says.

Maximum Drawdown = (Trough Value - Peak Value) / Peak Value

Both values are taken within the stated reporting period; the result is negative.

The workbook's own case, for the quarter ended 31 March 2020.

Fund PQR   MDD = (545.61 - 632.33) / 632.33  =  -13.71%
NIFTY 50   MDD = (7,610.25 - 12,430.50) / 12,430.50  =  -38.78%

In rupees, the fund gave back Rs 86.72 crore of AUM from its February peak. Read alone, that looks like a bad quarter. Read against a benchmark that fell 38.78% over the same window, it says the manager took roughly a third of the market's downside.

That comparison is the entire lesson. Change the reporting period to the full year and both numbers change; drop the benchmark column and the fund's number cannot be judged at all.

Now the forward-looking view. If the same fund reports a 95% one-month VaR of Rs 12 crore on a Rs 545 crore AUM, it is saying there is a 95% chance of losing no more than about 2.2% in a month. It is not saying the loss in the remaining 5% of months is Rs 12 crore — March 2020 is exactly the month that lives in that tail, and VaR is silent about its size.

Chapter 9 defines downside risk in the footnote to section 9.6.3 (maximum drawdown) and returns to it in section 9.9.3, where value at risk is introduced as a measure used to quantify the downside risk of an AIF. Expect an MDD computation from a table of AUM values, a benchmark comparison, and the standard VaR interpretation sentence.

• MDD is period-dependent. Change the reporting window and the peak, the trough and the answer all change.
• A large MDD is not automatically bad. The workbook's own conclusion is that -13.71% was good, because the benchmark fell -38.78%.
• MDD says nothing about frequency or recovery time — only about the depth of the worst fall until a new peak.
• VaR at 95% does not cap the loss. It is silent on the size of the loss in the 5% tail; that is its stated limitation.
• Standard deviation measures both tails; downside risk is only one of them, which is why the two can rank funds differently.
• Peak and trough are read off AUM for a fund and off index values for a benchmark — do not mix a fund's NAV per unit with its AUM in the same computation.