Volatility differences

Here is the simplest implementation of an implied volatility estimation in VBA by Espen Haug, it is easily portable to Python, however, I am not sure, but I think QuantLib in Python has a built-in Implied volatility estimator. Here is the code for reference:

Public Function GBlackScholesImpVolBisection(CallPutFlag As String, S As Double, _
            X As Double, T As Double, r As Double, b As Double, cm As Double) As Variant

Dim vLow As Double, vHigh As Double, vi As Double
Dim cLow As Double, cHigh As Double, epsilon As Double
Dim counter As Integer

vLow = 0.005
vHigh = 4
epsilon = 0.00000001
cLow = GBlackScholes(CallPutFlag, S, X, T, r, b, vLow)
cHigh = GBlackScholes(CallPutFlag, S, X, T, r, b, vHigh)
counter = 0
vi = vLow + (cm - cLow) * (vHigh - vLow) / (cHigh - cLow)
While Abs(cm - GBlackScholes(CallPutFlag, S, X, T, r, b, vi)) > epsilon
    counter = counter + 1
    If counter = 100 Then
        GBlackScholesImpVolBisection = "NA"
        Exit Function
    End If
    If GBlackScholes(CallPutFlag, S, X, T, r, b, vi) < cm Then
        vLow = vi
    Else
        vHigh = vi
    End If
    cLow = GBlackScholes(CallPutFlag, S, X, T, r, b, vLow)
    cHigh = GBlackScholes(CallPutFlag, S, X, T, r, b, vHigh)
    vi = vLow + (cm - cLow) * (vHigh - vLow) / (cHigh - cLow)
Wend
GBlackScholesImpVolBisection = vi

End Function

The black scholes equation needs to be put into a function to be called as a subroutine of this function.

The most common methods apart from the standard deviation of returns which is the most common method of estimating volatility are, Parkinsons extreme value method Sheldon Natenberg suggests

These are all Historic or realized volatility estimates NOT Implied volatility which is a whole different ball game.

I utilise R not python, but the math is the same first:

Then take the standard deviation of these, however to get a stable picture over time utilise GARCH estimation, and a great way to proxy implied volatility which is the key to statistical arbitrage, is exponentially weighted moving average volatility estimation. You will be set on a journey of deep discovery with these topics and will make yourself a more profitable and better trader.

There are two big problems with what you are doing.

First, you are trying to estimate the standard deviation of prices instead of price changes. Prices are not stationary: wait long enough and they are likely to head off to 0 or a very large number; and, they don't tend to stay around a certain value. You cannot reliably estimate parameters using just non-stationary data.

You could instead look at price changes. That is better, but it runs into your second problem: price changes for assets with high prices tend to be greater than price changes for assets with low prices.

The best way to handle this is to work with log-returns: differences in log(prices). This also eliminates some mechanical skewness that you get if you use standard returns. The standard deviation of, say, daily log-returns gets you a daily volatility. Scale that up to an annual volatility (what is typically quoted) by multiplying by sqrt(T) where T is the number of trading days in a year.

That's why you should measure standard deviation of returns.

Let me expand a bit your example:

prices_currency_1 = [1, 100 120]
prices_currency_2 = [.1, 10 12]

Returns:

returns_currency_1 = [ 9900% 20%]
returns_currency_2 = [ 9900% 20%]

So as you can see, the volatility of the currency itself, it seems that the first one is more volatile. But in terms of returns, which is what we care, the volatility of the two currencies is the same.

That's why you do not compare the volatility of Tesla and Berkshire stock price for example. You compare the volatility of their returns.

Another way of saying this, is that the volatility of investing 1 dollar in currency 1, or 1 dollar in currency 2 (the first case you buy 1 unit the second case you buy 10 units), is the same.