I was rather suprised to find out the following result from our model: -The cost of an Average Rate Option (fx in my case) increased with the number of times observations were taken. So a 1-year AVR with 220 (taking an average of the price on every trading day) counting periods actually cost more than one with 12 (observe prices once a month). This runs counterintuitive to me, since the more observations you take the lower the volatility should be. Apparently there exists some unbeknown to me hitch . Thanks in advance!

Obvious tests to try:1. One datapoint, at expiry, should be vanilla2. Two data points, one at start and one at end, should be 0.5 x vanilla with new strikeP

220 trading days per year? That's a lot of holidays (there are 260 or 261 weekdays in a year). Check your volatility adjustment.One reason the daily averaging can give a higher option price is the volatility measured daily will almost always be greater than the monthly volatility scaled to the same time period. If this is actually true, that is if the daily average is more volatile than the monthly average due to mean reversion, bid/ask spreads, or something else; the daily average could really be a more valuable option.But under standard continuous Black-Scholes assumptions, the more frequent observations should lower the option value slightly.

Average Rate Call withStock 100Vol 0.2Int rate 0.05Strike 100Expiry 1Simple MC simulation (10,000 runs) gives:N / Value101 / 5.6151 / 5.8321 / 5.7711 / 5.726 / 5.533 / 5.472 / 5.27where N is number of samples (inc. one at start and one at end). ?P

Thanks for your suggestions. Let me shore up a real example (with 249 days in a year USD/EUR 1-year AVR -spot .8780 -strike .8724 (atmf for 1 year) -fwdpts -0.00559 -dom rate 2.54 -vol 11.62TV vanilla - 4.4903%Total observations TV Exotic-------------------------------249 2.7531%100 2.749350 2.742725 2.729710 2.69015 2.62272 2.40831 4.4903So there you have it. I don't have a way to set when observations are taken - i assume they are made to split the time interval into equal periods.Would anybody happen to have an AVR model handy to see how the number of observations figures in the formula?