mod12lec59

IIT Roorkee July 201
61 followers
Jan 10, 2024
Value at Risk: Definition & Computation
Welcome back, so in the last lecture I introduced the concept of value at risk, and let us go
back and recap a bit, given a portfolio W for which we are to compute the value at risk
and a certain probability which we call the confidence level and a Time Horizon, these
are the three inputs that go into the value at risk, value at risk attempts to answer
the question.
I am p percent certain where p is the confidence level, I am p percent certain that the loss
on my portfolio W, the loss of my portfolio W shall not exceed that number what we call
the value at risk let us call it D over a period of T days which is the time Horizon.
So basically we are trying to answer this question I am p percent certain that my portfolio
W will not suffer a loss which is greater than the value at risk over a period of T
days subject to this probability p that we call the confidence level.
In other words you can also interpret it as the loss level which could be exceeded with
the probability of 1 minus p, you could also interpret it as given a set of say a 100 observations
or given a set of 100 days data it would only be in 1 minus p occasions that the loss on
the portfolio would exceed the value at risk, right?
So if the underlying distribution is normal, then this is my, this whole area represents
my confidence level this entire area represents my confidence level and given this confidence
level which is predetermined, which I have to choose which is an input to the model,
I work out the value at risk as the maximum (possible) maximum loss, or the maximum expected
loss subject to this probability. So given this probability my maximum expected
loss is going to be this value of the loss or the change in price. Loss is basically
the change in price over the time Horizon, whatever we are talking about.
So, this particular quantity here will represent the value at risk loss at this confidence
level which is the probability covered by the un-shaded area and this shaded area will
obviously be 1 minus p. Now, there is another fallout of this immediate fallout of this
that if the confidence level increases, what does it mean? It means that this area will
increase say earlier the confidence level was 95 percent so this un-shaded area was
95 percent, now if the confidence level has you have changed the confidence level to 99
percent let us say then this area should be 99 percent.
This will push this particular point more to the left because the right hand side area
should now be 99 percent instead of 95 percent of the entire area and therefore in other
words, what will happen is the value at risk will increase in absolute terms and value
of risk is obviously a negative quantity and it has to be it is usually represented as
the absolute value of the negative quantity. So value at risk in terms of magnitude will
increase given an increase in the confidence level. Now, we come to the calculation of
value at risk, now calculation of value of at risk can be done in two fundamental ways,
either we can use a historical approach or an empirical approach or we can work out the
value at risk assuming a certain distribution, assuming a certain distribution a model for
the price of the portfolio or price of the asset for which we are to determine the value
at risk. We will talk about both of them. Let us start with the historical approach.
Let us say these are the assumptions that go into the historical approach the portfolio
under consideration does not change over the forecast horizon and please note when we compute
value at risk the forecast horizon is usually very small we initially calculate the value
at risk over a period of 1 day and if required, if required we extrapolate it over T days
by using the square root rule. The historical data used contains information
useful for forecasting the loss distribution that is it is symptomatic of what is likely
to occur in the future and the third is that the historical data follows a specific distribution,
these are the assumptions that go into the computation of value at risk by the empirical
method.
Now, let us go to the technicalities. Let us assume for simplicity that I have a portfolio
and that portfolio is a function of a certain index for simplicity let us let us assume
for concreteness let us assume that that particular index is the BSE Sensex, S and P BSE Sensex.
In other words, the value of my portfolio is in some way related to it has some functional
dependence on the S&P BSE Sensex, it is a one index relationship and my portfolio value
is captured by value of the Sensex. So, let us let us again for the sake of concreteness
work out a one day where 99 percent value at risk for this particular portfolio and
for this purpose let us assume that we have collected 501 units of data, 501 units of
data going back in time, say today is t equal to 0, from t equal to 0 to t equal to minus
500, minus 500 means we are going back in time, we are collecting historical data, so
data for today, yesterday, day before yesterday and so on, we continue up to 500 days going
backwards. So the earliest data will be for t equal to minus 500 and the latest data that
we have is for T equal to 0, there will be 501 units of your data in respect of values
of the Sensex that is what is the input to the model.
Now, what we do is in this particular case, what we do is the first step is that we construct
scenarios or we create scenarios see know the Sensex value as of today so let us call
today as t equal to 0. So we know the Sensex value at t equal to 0, what we are worried
about, what we want to work out is to work out the Sensex value at t equal to 1 that
is tomorrow and once we know or once we project the Sensex value at t equal to 1, we are able
to capture or we are able to determine the value of our portfolio corresponding to the
projected value of the Sensex. And therefore we are able to calculate the
change in value of our portfolio corresponding to that projected value of this Sensex and
therefore we are able to because we will have a number of such values of this Sensex, number
of such projections how? we will come back to, but we will have a number of such projections
and therefore we are able to define a distribution of the changes in value of this our portfolio
between t equal to 1 that is tomorrow and today which will ultimately enable us to determine
the value at risk that is the underlying mechanism that is to be followed.
So let us now start with this, so we have got 501 days data, the first step is yeah
I have a blank slide here, so what we start with is 501 days data. Let us say today is
t equal to 0, tomorrow is t equal to 1 and this is 500 days data, 501 days data this
is t equal to minus 500 and I have data Sensex values data is Sensex values I have Sensex
values from t equal to minus 500 to t equal to 0 and let us call those Sensex values,
let us give them a name let us call them M minus 500, M minus 499 and so on and here
it will be M 0, right. So, what we do now is in order to project
the value of the Sensex at M1 at t equal to 1 you can call it M1 or you can call it scenario
1 this is also equal to scenario 1, in order to project the value at t equal to 1 the first
scenario that we work out is by assuming that the change in value between today and tomorrow,
the percentage change in value between today and tomorrow that is, this is scenario 1 let
us call it, M1 upon M0 the percentage change in value between today and tomorrow that is
M1 upon M0 is equal to the percentage change in value between the last observation that
we have that is M minus 500 and the next earlier observation which is M minus 499.
So we want that this change in value should be equal to M minus 499 divided by M minus
500, this will give me the first scenario, the first scenario value of M1 or in other
words scenario 1 for M1 will be equal to M0 into M minus 499 divided by M minus 500. So
this is the way I will get the first scenario. Now, what about the second scenario? The second
scenario I will obtain similarly. What is the second scenario? The second possible value
that I feel would be of the Sensex tomorrow given whatever information I have about the
data that is the history of this Sensex. So, S2 will be equal to, now in this case
what I assume is that the change in value percentage change in value between tomorrow
and today M1 upon M0 is equal to the percentage change in value between M minus 498 divided
by M minus 499 and this way we will work out, this is the second scenario, the second possible
value of the Sensex that I perceive to be viable at tomorrow. Similarly I will work
out the third scenario, fourth scenario, fifth scenario and so on right.
So we worked out scenario 1, scenario 2 and the various scenarios by using the chain of
values that we had collected for the 500 days of Sensex and we arrive at 500 possible scenarios
for the projected values of Sensex as of tomorrow given the value that we have as of today.
As you can see the first Sensex projection or the first value of first value of the scenario
for Sensex as of tomorrow is given by M0 into M minus 499 upon 500.
And similarly, the second scenario is given by M0 into minus 498 upon minus 499 and similarly
for the rest, so will have 500 possible Sensex values, possible Sensex values for tomorrow
based on the chain of data the historical chain of data 501 observations of the historical
data that we have inputted into this particular model.
Now, what we do is for each scenario knowing the functional dependence of our portfolio
on the Sensex value for each scenario we work out the value of the portfolio (V,) V1 is
equal to F of S1, S1 is scenario 1 of the Sensex, V2 function of S2 Sensex and they,
say for example, if you have the CAPM model based functional relationship in your portfolio,
you can use this expression which is here, which is here and otherwise of course whatever
that functional relationship is the bottom line is that we have to work out the value
of the portfolio corresponding to corresponding to each value of the Sensex, right.
So, now we have 500 possible values of our portfolio, we have translated the 500 possible
values of the Sensex that we projected for tomorrow into 500 possible values of our portfolio
for tomorrow knowing each of these 500 values we work out the changes in value of our portfolio
changes in value of our portfolio between today and tomorrow.
We know the value of our portfolio today, we have now worked out various scenarios for
the value of a portfolio tomorrow and therefore we can work out the change in value corresponding
to each scenario. What we do now is will plot a probability distribution or we plot a histogram,
whatever how so ever we work out, but we work out the fifth because we are talking about
99 percent value at risk and we are talking about 500 units of data.
So what we will be looking at if we arrange the changes in value in descending order the
best, the maximum positive change in value at the top and the maximum negative change
in value at the bottom, negative change in value represents a loss, a positive change
in value represents a profit. So if we arrange the data arrange the changes
is value of our portfolio as given in our scenarios in descending order, the fifth worst
possible loss the 99th percentile will give us the 99 percent value at risk for one day.
If you want to extrapolate it over T days you multiply by root T. So it is as simple
as this if you want to use historical data for the computation of value at risk.
This is an example that we have collected sample hypothetical values.
These are hypothetical values for various days of the Sensex and based on that we have
worked out.
These scenarios, first scenario the Sensex value comes out to be this, second scenario
Sensex value comes out to be this.
The working is shown in this slide, value at Sensex on September 20, which is this is
my t equal to 0, this is my t equal to 0 1 1, 0 2 2, this is my t equal to 0 and this
is August 7 2 0 1 6 is t equal to minus 500 August 8 is t is equal to minus 499 please
note this so knowing these two values the percentage between them will be equal to the
percentage between today change between today and tomorrow and thereby we get this value.
We get this value for the change in for the Sensex for t equal to 1 that is tomorrow and
similarly the various other scenarios are worked out for Sensex and then we work out
the portfolio value corresponding to each scenario and we work out the value at risk.
Now what happens, what happens if the functional dependence of our portfolio is not confined
to a single index? For example our portfolio may not be simply
mimicking the S and P BSE Sensex, but there may be some content of foreign investment
in our portfolio, there may be some content of may be foreign currency in our portfolio
and therefore there would be different indices which would be which would be trailed by or
which would be reflected or which would be mimicked by different constituents of our
portfolio. Let us assume that our portfolio W is a given
by this expression a weighted combination of the functional dependence of different
indices M1, M2, M3 these are different indices, BSE Sensex maybe the New York stock index
and maybe the financial times index of the UK and so on maybe the Nikkei index of the
of Japan and so on. So suppose it is a heterogeneous portfolio,
but we can identify the components or even if it is not there, there is some functional
relationship between the various components and the indices of the portfolio. The basic
thing is that we must be able to determine the value of the portfolio given these various
indices. Suppose I know the values of certain indices,
I must be able to ascertain the value of the portfolio once if that is available if that
information is available if that functional relationship is available, then we can simply
extrapolate what we did for the for the single index portfolio to a many index portfolio
by repeating the exercise, extrapolating the exercise which we did for the single index
portfolio. We collect data for a sufficient number of
historical days, a sufficient number of days in the past on the basis of which we work
out scenarios for each of these indices each of these indices and for each scenario for
each of these indices for each scenario for each of these indices we work out the portfolio
value because we know the functional relationship. So we will have the portfolio value corresponding
to each of these scenarios and knowing the portfolio value for each of these scenarios
we are then able to fit a probability distribution or we are able to the arrange the changes
in portfolio values that is todays value vis a vis tomorrows value, tomorrows value minus
todays value is rather, tomorrows portfolio value in each scenario minus todays portfolio
value we arrange them in descending order and we take the appropriate percentile of
that of that series to arrive at the value at risk. So it is simply an extrapolation
of what we did for the single variable where we but we need data for each of those variables
ofcourse, the rest of the exercise absolutely parallel.
So that done we now move to calculation of value at risk using models. Let us start with
the simplest case single asset case. Now for the single asset case it is very common, it
is very the common practice to use the Black-Scholes model.
And because we are talking about one day in general we determine the value at risk for
one day and then extrapolate the infinitesimal Black-Scholes model is good enough. So we
use this model dS is equal to Mu as D dt plus sigma as DW and DW can be written as Z under
root dt, we have talked about it in lot of detail when we discuss Brownian motion. So
dS is equal to Mu into S dt plus Sigma into s Z under root dt.
If we are talking about one day and if all these parameters are for one day, we can take
dt equal to 1 and therefore our expression simply or model simplifies to dS the change
in value over one day change in value of our stock or for asset over one day is equal to
Mu per day into S, dt is 1 plus sigma per day into S into Z. Obviously the expected
value of the change, now what is dS? DS is tomorrow stock price minus today stock
price, S1 minus S naught, so E of dS per day is equal to (Mu into S day Mu into) Mu day
into S Plus standard deviation day, per day that is given by sigma day into S. Now it
is conventional, it is not really conventional value, but it is an approximation but an approximation
which is usually employed and which is to some extent justified.
The assumption is you see the point is if you look at this the Black-Scholes model you
will clearly see that the expected return scales as T, expected return over a time dt
is given by Mu dt so expected return scales as t. However, if you look at the standard
deviation the standard deviation scales as root dt, I repeat the expected return scales
as t or dt for that matter and the standard deviation scales us root dt or root t, whatever
the case may be. Now, the important thing is if that is so
when we convert the annual return and the annual volatility into a daily volatility
or a daily return we divide by the number of days in the year or the number of trading
days in the year as far as the return goes and when we work out the volatility per day,
we divide this by the square root of the number of days or the number of trading days. The
result is, the result you can see is let me give you an example.
The result is suppose 20 percent is the annual return and 20 percent is the annual volatility
and we are supposed to work out the daily return and the daily volatility then 20 percent
per annum corresponds to a daily return of 0.08 percent and 20 percent per annum corresponds
to a daily volatility of 1.26 percent. Look at the gigantic difference between the two,
the return per day corresponding to the same corresponding to the same annual return is
0.08 percent and the standard deviation or the volatility per day is 1.26 percent, so
the Gap is gigantic. That being the case it is fair enough that
we ignore the first term here and we retain the second term and we write that dS per day,
dS per day is equal to Sigma per day into S into Z and we ignore the first and the Mu
term we ignore that because mu per day is going to be pretty small. That being the case,
in other words what we are saying is what we are simply saying is that the expected
change in price expected change in price on a daily basis on a per day basis is negligible,
but the standard deviation of the change in price is given by Mu per day into S because
the standard deviation of Z is 1. In other words, we can extend this and say
that dS per day is normally distributed with a mean of 0 because we are assuming that Mu
per day is zero or close to zero. So dS per day is normally distributed with a mean of
0 and a variance of sigma day into S (where) whole square sigma day into S whole square.
Now, let us let us go to the modalities of value at risk, this was the basic model. Now,
given a confidence level p, now this is the confidence level p, this is the probability
p given a confidence level p we define value at risk as the “least” I put least in
inverted commas because please note value at risk is the maximum loss, maximum loss
is tantamount to minimum profit. You see normally we work out the change in price by tomorrow’s
price minus today’s price at T minus price as 0, change in price is determined in this
manner. Therefore, if tomorrow’s price is more than
today’s price you make a profit and if tomorrow’s price is less than today’s then you make
a loss. So when we are talking, in other words, if dS is positive you are making a profit,
if dS is negative you are making a loss. So because we are talking about the maximum loss,
it could be the least value of dS because we are maximum loss amounts to minimum profit
or minimum change in this expression. The (negative) see negative included negative
included it amounts to the minimum change in dS or change in price.
Therefore, we can define value at risk as that least value of dS, that least value of
dS such that the probability of dS being greater than that, dS been greater is good that is
profit or at least the lesser loss so dS per day should be greater than that minimum change
which is the maximum loss. This is the minimum change, this is also the maximum loss, dS
star is the minimum change it is also the maximum loss and this is my value at risk
such that the probability is equal to p this whole area un-shaded area is equal to p.
I repeat, why we are using least? Because we are talking about maximum loss and maximum
loss is nothing but minus of profit and we want the profit to be least, we are talking
about a profit distribution, we are talking about the gain distribution we are talking
about S1 minus S0, we are not talking about S0 minus S1, we are not talking about S0 minus
S1, we are talking about S1 minus S0 and therefore positive dS will mean profit, negative dS
will mean loss and therefore we want negative as much as possible subject to this probability
and we are happy with the negative being less than this quantity, which is what it is representing,
negative being greater than this quantity this portion, we are happy with we do not
want this portion. So, but dS per day is normally distributed
as we did in the previous slide normally distributed with a mean of zero and a standard deviation
of sigma per day into S, therefore if Z is the standard normal variate because the mean
is 0 we have Z is equal to dS per day upon sigma day into (s) this is my standard deviation,
this is this is my what you call X minus X bar, X bar is 0 so X minus X bar upon sigma,
this is my Z which gives me dS per day is equal to Z into sigma day into S.
We can write, we can therefore write this expression as p is equal to probability dS
per day is greater than, is greater than dS star per day, which is my value at risk, this
is my value at risk, dS per day is greater than dS star which is value at risk or putting
these two values we get p Z sigma per day into S greater than Z star sigma day into
S this simplifies to p, small p is equal to probability of Z being greater than Z star.
So this is this point corresponds to Z Star, this point corresponds to Z star and such
that and how do we determine Z star? You see, we try to understand the point P is probability
P is given to us this is the confidence level given P we can find Z star from the normal
distribution table such that P Z greater than Z star is equal to P.
So Z Star can be formed from the normal distribution, because it will correspond to this point and
therefore Z star, knowing Z star we can calculate value at risk, how? This value at risk is
equal to Z star, sigma day into S, this Z confidence is nothing but Z star, sigma day
into S, so and if you want a T day value at risk you multiply by root T, so this whole
expression gives me the T day value at risk for a at a confidence level which corresponds
to Z confidence. Right, so do we, now just a bit of explanation about the root T rule
which I just mentioned in a few minutes ago.
You see, again we come back to this using this per day parameters Mu per day and S per
day we can write dS over T days, dS over T days as Mu per day into S into dt sigma per
day into s into Z into root dt where dt is now measured in days. Please note this because
we are using Mu in days, so we have to use dt also in days it has to be consistency.
And that that gives me that the standard deviation of dS over T days is equal to sigma day into
S into root dt which clearly shows which clearly shows that my standard deviation is scaling
in terms of the square root of time. Whereas, the expected value of S is scaling
or dS scaling in terms along or with time with t this is proportional to T in a sense
and this is proportional to root T, that is why we had that situation where we could ignore
this particular quantity the first factor and we could simply model the change in stock
price over a very small period of time by this second factor alone.
Let us do an example, consider a position consisting of a dollar 0.1 million investment
in an asset A. Assume that the annual volatility of the asset returns is 30 percent there are
252 trading days, what is the 5 day 99 percent value at risk for this portfolio?
Amount of investment is 0.1 million U.S. dollars, sigma is given as 0.30 that is 30 percent
trading days are 252, the Z value corresponding to 99 percent confidence level is 2.33 this
can be obtained from the normal distribution table, T is equal to 5 days therefore the
value at risk is given by 2.33 into this S into 0.30, 0.30 divided by root 252 gives
me 0.30 divided by root 252 gives me the daily volatility. So that is where this root 252
comes 0.30 root 252 is the daily volatility, root 5 because we are working out the value
at risk over 5 days and if you simplify this expression you arrive at this quantity 9846
dollars over 5, this is the value at risk over 5 days.
Now (we simply) a straightforward extension to a portfolio comprising of two assets, variance
of a pair of assets X and Y is given by this expression please note there is one important
point here, otherwise the formulas is quite forward it is very common we use it to invariably
in portfolio analysis. The important point is this left hand side variance is in terms
of money unit so it will be money unit squared, this S is in money units, sigma is in percentage
on the right hand side, please note this, left hand side sigma is in money units and
the right hand side S is in money unit, sigma is in fractions not percentage divided by
100, this is also in money unit, this is in percentage divided by 100 and row is of course
dimensionless. So S is in money, this is also in money, sigma
X is in percentage divided by 100, sigma Y is in percentage divided by 100. So this procedure
is quite straight forward I illustrated with an example.
Consider a position consisting of dollar 0.1 million in asset A and 0.1 million investment
in asset B, assumed that the daily volatilities of both the assets are 1 percent and that
the coefficient of correlation between returns is 0.3. What is the 5 day 99 percent value
at risk of this portfolio? 5 day, 99 percent.
Now, in this case SA is equal to SB is equal to 0.1 million sigma A Sigma B is equal to
1 percent that is equal to 0.01. So the variance of the portfolio daily change is given by
this expression which we had on the previous slide, if you put the respective values and
you take the square root of the left hand side we get sigma A plus B this is in money
terms now in US dollars that is 1612.45 this is the daily change of the portfolio daily
standard deviation of the portfolio, of the portfolio consisting of both the assets A
and B. Therefore the standard deviation of a 5 day change in price is multiplied by root
5 and the Z value corresponding to a 99 percent confidence is 2.33 minus 2.33, but we ignore
the minus sign when we talk about value at risk.
And therefore, the 5 day 99 percent value at risk works or 2.33 into (3 point) 3605
that is 8401, from where do we get 3605? Well, we get it from 1612 into root 5 this is equal
to 3605, we will continue after the break, thank you.
