Hello, Hope someone can help me with this. I have a time series of daily performance data going back several years. How would I go about calculating the 1-day, 10-day, and 30 day VAR at a 99% confidence level? Do i just simply calculate it using the last 1, 10, and 30 days, or is there another preferred method (e.g. taking an annualized figure and deriving a 1, 10, and 30 day average)? Any help is very much appreciated!
several different methods: 1) calc daily sample st.dev from the daily series then convert the daily st.dev to 10 day (multiply the daily by root 10), and 30 day (multiply the daily by root 30). Then if you want the 99% confidence level, subtract 2.3 st.devs from the expected returns over 1, 10, and 30 days. Multiply this return (loss) by the amount at stake. 2) since daily returns may have serial correlation, you could calculate average 10 day and 30 day returns over the period, then calc sample st.devs for 10 days and 30 days from the 10 day and 30 day av returns. Then apply the confidence level (2.33 st,devs below expected return. 3) personally if I were using several years of data I would use the historical data to calc a sample annual or quarterly st.dev, then convert that to a 1 day, 10 day and 30 day st.dev. (to minimise serial correlation). Then apply the confidence level to each. (these 3 are variations of the Analytical method. The results would be very similar - but remember that VaR is just a broad estimate - a very blunt tool - not much good by itself - riddled with assumptions and conditions) 4) you could rank the daily returns - eg if you have 10 years of daily data, that gives 2,500 daily returns. The minimum loss at the 99% confidence level would be the 25th worst return. Similar rankings for the 10 day and 30 day returns. (this is the Historical method) 5) Monte Carlo simulation - run simulations of outcomes based on the historical returns and st.devs, then rank the outcomes. If you have 1,000 outcomes, the 99% confidence level would be the 10th worst result. is that what you were getting at?
Perfect! Thank you so much. I actually had the analytical method in mind when I asked, but good also to know how its done using the other methods.
or you could pick a number, double it, then add your house number - that’s what most CFOs do and the Board just says “that’s nice - pass the scotch!” VaR is dead - after 2008… McKinsey, BCG, etc will come up a new “next best thing” and charge millions implementing it all over the world - just in time for the next crash!
The thought did cross my mind and I just might revert to it yet. I wish you could convince some individuals I’ve come across on VAR being dead. I fully agree, though!