PawelIn today’s part in the series, we will discuss the concept of Incremental VaR (Value-at-Risk) and its impact on portfolio.
Previously, in Part 1 we implemented VaR analytical methodology to an exemplary 7-asset portfolio of equities deriving a running portfolio VaR, marginal VaR, and component VaR. In Part 2 we have examined an approximative impact of historical data on the covariance matrix estimation used in the portfolio VaR calculations.
Altitude Adjustment, i.e. Incremental VaR
For a current portfolio holding $N$ assets with a total exposure of $C$ (dollars) we get a 95% portfolio VaR estimation simply computing:
$$
\mbox{VaR}_P = 1.65 \sigma_P C = 1.65\left(\boldsymbol{d}’ \boldsymbol{M}_2 \boldsymbol{d} \right)^{1/2}
$$ This can be estimated any time. For example, if we evaluate a portfolio of US stocks on the daily basis, and new data are available to be downloaded after 4pm when markets close, 1-day VaR can be recalculated. If it increases day-over-day and, say, one of component VaR behaves in an abnormal way (rising a warning flag), we may want to do something about that particular position: remove it completely or decrease the exposure.
A second scenario could be more optimistic. Say, we have extra cash to invest and we want to increase one (or few) currently held positions in our portfolio. Adding more shares of $i$-th asset in portfolio will increase portfolio VaR. The secret is in not pushing running engines too much and keep the oil temperature within acceptable brackets.
Alternatively, as a risk manager working in a global bank, the bank’s portfolio lists 10,453 open positions (across various markets and continents) and you have an incoming proposal of a new trade coming from one of the bank’s client. He wants to invest $\$$2,650,000 in highly illiquid asset. It is physically possible but the bank VaR may increase significantly over night. How much?
Well, in this or any similar scenarios, you need to perform the portfolio revaluation employing the incremental VaR procedure. Your aircraft may be brand new but remember that big planes fall from the sky too.
In general, by the incremental VaR we consider the difference between a new portfolio VaR (including (a) new trade(s) or a change in (a) current holding(s)) and recent portfolio VaR:
$$
\mbox{iVaR}_P = \mbox{VaR}_{P+a} – \mbox{VaR}_P
$$ where $a$ symbolises the proposed change. The incremental VaR changes in a non-linear way as the position size can be increased/reduced in any amount. That is the main difference as compared to the marginal VaR.
It is possible to expand $\mbox{VaR}_{P+a}$ in series around the original point, ignoring second-order terms if the deviations $a$ are small:
$$
\mbox{VaR}_{P+a} = \mbox{VaR}_{P} + (\Delta \mbox{VaR})’a + …
$$ and by “small” I mean a dollar change in (a) current holding(s) much much less than $C$ under market exposure. In such case,
$$
\mbox{iVaR}_P^{\rm approx} \approx (\Delta \mbox{VaR})’a
$$ and the apostrophe denotes the transposition of the vector. This approximation becomes important for really large portfolios (e.g. in banks) where revaluation and a new estimation of bank VaR needs to be derived quickly. A time saver at a reliable precision of landing. In all other cases, a new vector of extra exposures can be added to present position sizes and reevaluated. Let’s see how it works in practice!
Case Study A: Increasing Position of One Asset
Let’s recall the same initial settings as we have done in Part 1. We deal with 7-asset portfolio of US equities as of 12th of January, 2015. There is nearly 3 million dollars invested in stock market (I wish it were mine! ;) and based on 3-year historical data of all underlying stocks, the estimated 1-day 95% portfolio VaR is $\$$33,027.94 as forecasted for the next trading day. Id est, there is 5% of chances that we can lose 1.1% of the portfolio value before English men in New York will rise their cups of tea:
% Applied Portfolio VaR Decomposition. (3) Incremental VaR and Portfolio
% Revaluation.
%
% (c) 2015 QuantAtRisk.com, by Pawel Lachowicz
clear all; close all; clc;
format bank
%{
% Read the list of the portfolio components
fileID = fopen('portfolio.lst');
tmp = textscan(fileID, '%s');
fclose(fileID);
pc=tmp{1}; % a list as a cell array
% fetch stock data for last 3 years since:
t0=735976; % 12-Jan-2015
date2=datestr(t0,'yyyy-mm-dd'); % from
date1=datestr(t0-3*365,'yyyy-mm-dd'); % to
% create an empty array for storing stock data
stockd={};
% scan the tickers and fetch the data from Quandl.com
for i=1:length(pc)
quandlc=['WIKI/',pc{i}];
fprintf('%4.0f %s\n',i,quandlc);
% fetch the data of the stock from Quandl
% using recommanded Quandl's command and
% saving them directly into Matlab's FTS object (fts)
fts=0;
[fts,headers]=Quandl.get(quandlc,'type','fints', ...
'authcode','YourQuandlCode',...
'start_date',date1,'end_date',date2);
stockd{i}=fts; % entire FTS object in an array's cell
end
save data
%}
load data
% limit data to 3 years of business days, select adjusted
% close price time-series and calculate return-series
rs=[];
for i=1:length(pc)
cp=fts2mat(stockd{i}.Adj_Close,0);
rv=cp(2:end,1)./cp(1:end-1,1)-1;
rs=[rs rv(end-(3*5*4*12):end)];
end
rs(isnan(rs))=0.0;
% covariance matrix
M2=cov(rs)
% an initial dollar expose per position
d=[ 55621.00; ...
101017.00; ...
23409.00; ...
1320814.00; ...
131145.00; ...
321124.00; ...
1046867.00]
% invested capital of C
C=sum(d)
% the portfolio volatility
vol_P=sqrt(d'*M2*d);
% diversified portfolio VaR
VaR_P=1.65*vol_P
% volatility of assets (based on daily returns)
v=sqrt(diag(cov(rs)));
% individual and undiversified portfolio VaR
VaRi=1.65*abs(v.*d);
uVaR_P=sum(VaRi);
% the portfolio betas
beta=C*(M2*d)/vol_P^2;
% the portfolio marginal VaR
DVaR=1.65*(M2*d)/vol_P;
% the component VaR
cVaR=100*DVaR.*d/VaR_P; % [percent]
cVaRd=DVaR.*d; % dollars
Now, we want to add $\$$10,000 to our exposure in Walt Disney Co. stock. As of Jan/12 2015 DIS closes at $\$$94.48 therefore we consider to buy $\$$10,000/$\$$94.48=105.8 shares on the next day. We are flexible so we buy at the market price of $\$$95.23 on Jan/13, i.e. extra 105 shares in DIS. Therefore we added $105\times \$95.23$ or $\$$9,999.15 to our portfolio:
% proposed changes to current exposure
da=[0; ...
9999.15; ...
0; ...
0; ...
0; ...
0; ...
0 ];
Of course we can assume a new portfolio VaR estimation by adding extra 105 shares of DIS. In real weather conditions, this number can be lower. Keep that in mind.
The remaining part seems to be easy to code:
% new portfolio VaR VaR_Pa=1.65*sqrt((d+da)'*M2*(d+da)) % incremental VaR iVaR=VaR_Pa-VaR_P; % incremental VaR (approximation; see text) iVaR_approx=DVaR'*da; % new component VaR ncVaR=100*DVaR.*(d+da)/VaR_P; % [percent] ncVaRd=DVaR.*(d+da); % dollars
followed by a new risk report on the impact of proposed changes:
fprintf('Current exposure:\t\t\t $%1s\n',bankformat(C));
fprintf('Current Portfolio VaR (diversified):\t $%1s\n',bankformat(VaR_P));
fprintf('Proposed changes in holdings:\n');
for i=1:length(d)
fprintf(' %s\t $%-s\n',pc{i},bankformat(da(i)) );
end
fprintf('New Portfolio VaR (diversified):\t $%1s\n',bankformat(VaR_Pa));
fprintf('Incremental VaR:\t\t\t $%1s\n',bankformat(iVaR));
fprintf('Incremental VaR (approximation):\t $%1s\n',bankformat(iVaR_approx));
fprintf('Change in exposure:\t\t\t %-1.4f%%\n',sum(da)/C);
fprintf('Component VaR:\n');
fprintf('\t Current\t\t New\t\t\tChange in CVaR\n');
for i=1:length(d)
fprintf(' %s\t%6.2f%% $%-10s\t%6.2f%% $%-10s\t$%-10s\n', ...
pc{i},cVaR(i), ...
bankformat(cVaRd(i)),ncVaR(i), ...
bankformat(ncVaRd(i)),bankformat(ncVaRd(i)-cVaRd(i)) );
end
In this case study, the report says:
Current exposure: $2,999,997.00 Current Portfolio VaR (diversified): $33,027.94 Proposed changes in holdings: AAPL $0.00 DIS $9,999.15 IBM $0.00 JNJ $0.00 KO $0.00 NKE $0.00 TXN $0.00 New Portfolio VaR (diversified): $33,084.26 Incremental VaR: $56.31 Incremental VaR (approximation): $55.84 Change in exposure: 0.0033% Component VaR: Current New Change in CVaR AAPL 0.61% $202.99 0.61% $202.99 $0.00 DIS 1.71% $564.13 1.88% $619.97 $55.84 IBM 0.26% $84.43 0.26% $84.43 $0.00 JNJ 30.99% $10,235.65 30.99% $10,235.65 $0.00 KO 1.19% $392.02 1.19% $392.02 $0.00 NKE 9.10% $3,006.55 9.10% $3,006.55 $0.00 TXN 56.14% $18,542.15 56.14% $18,542.15 $0.00
That’s the great example to show how the approximation of the incremental VaR works in practice. $\$$10,000 added to DIS is “small” as compared to $C$ of nearly 3 million dollars in the game. By adding 105 shares of DIS we didn’t move 1-day portfolio VaR significantly which is good. Let’s play again.
Case Study B: Increase and Decrease Exposure
From the initial risk report we learnt that TXN contributed most to the overall portfolio VaR. Let’s say we have extra $\$$126,000 to invest and we think that AAPL would be the next shining star. In the same time, we want to reduce our market exposure to TXN by selling stocks of the total worth of $\$$500,000. This operation would be denoted as:
% proposed changes to current exposure
da=[126000; ...
0; ...
0; ...
0; ...
0; ...
0; ...
-500000 ];
and
Current exposure: $2,999,997.00 Current Portfolio VaR (diversified): $33,027.94 Proposed changes in holdings: AAPL $126,000.00 DIS $0.00 IBM $0.00 JNJ $0.00 KO $0.00 NKE $0.00 TXN $-500,000.00 New Portfolio VaR (diversified): $25,869.04 Incremental VaR: $-7,158.90 Incremental VaR (approximation): $-8,396.16 Change in exposure: -0.1247% Component VaR: Current New Change in CVaR AAPL 0.61% $202.99 2.01% $662.85 $459.85 DIS 1.71% $564.13 1.71% $564.13 $0.00 IBM 0.26% $84.43 0.26% $84.43 $0.00 JNJ 30.99% $10,235.65 30.99% $10,235.65 $0.00 KO 1.19% $392.02 1.19% $392.02 $0.00 NKE 9.10% $3,006.55 9.10% $3,006.55 $0.00 TXN 56.14% $18,542.15 29.33% $9,686.13 $-8,856.02
The incremental VaR reacts to a massive sell out of TXN reducing the overall portfolio VaR significantly. Adding $\$$126k to APPL (share number precision omitted for the sake of simplicity) does not affect the portfolio risk levels in any dramatic way.
Case Study C: Adding Two New Assets
Your are a child of a new era, the era of credit cards. More and more people use them worldwide. You have this brilliant idea to add to your portfolio extra $\$$100,000 invested in (split equally between) Mastercard Inc. and Visa Inc.
Considering such an operation requires a new portfolio revaluation. Let’s do it step by step. First, let’s fetch historical data for both MA and V stocks, the same way as we did it at the very beginning. We save the tickers names in a separate file named proposal.lst:
MA V
We supplement out Matlab code with extra few lines:
% Read the list of the portfolio components
fileID = fopen('proposal.lst');
tmp = textscan(fileID, '%s');
fclose(fileID);
pcE=tmp{1}; % a list as a cell array
% fetch stock data for last 3 years since:
t0=735976; % 12-Jan-2015
date2=datestr(t0,'yyyy-mm-dd'); % from
date1=datestr(t0-3*365,'yyyy-mm-dd'); % to
% create an empty array for storing stock data
stockd={};
% scan the tickers and fetch the data from Quandl.com
for i=1:length(pcE)
quandlc=['WIKI/',pcE{i}];
fprintf('%4.0f %s\n',i,quandlc);
% fetch the data of the stock from Quandl
% using recommanded Quandl's command and
% saving them directly into Matlab's FTS object (fts)
fts=0;
[fts,headers]=Quandl.get(quandlc,'type','fints', ...
'authcode','YourQuandlCode',...
'start_date',date1,'end_date',date2);
stockdp{i}=fts; % entire FTS object in an array's cell
end
save datap
%}
load datap
% limit data to 3 years of business days, select adjusted
% close price time-series and calculate return-series
rsp=[];
for i=1:length(pcE)
cp=fts2mat(stockdp{i}.Adj_Close,0);
rv=cp(2:end,1)./cp(1:end-1,1)-1;
rsp=[rsp rv(end-(3*5*4*12):end)];
end
rsp(isnan(rsp))=0.0;
As quickly as new data arrive, we build and extend return-series matrix, now not $721×7$ but $721×9$ in its dimensions:
rsE=[rs rsp];
What follows is pretty intuitive in coding:
% covariance matrix
M2E=cov(rsE);
% proposed changes to current exposure
daE=[0; ...
0; ...
0; ...
0; ...
0; ...
0; ...
0; ...
50000; ...
50000 ];
% new portfolio VaR
d=[d; 0; 0] % we have to extend it to allow d+daE
VaR_PaE=1.65*sqrt((d+daE)'*M2E*(d+daE))
% a running list of tickers, too
pcE=[pc; pcE]
% incremental VaR
iVaRE=VaR_PaE-VaR_P;
fprintf('Current exposure:\t\t\t $%1s\n',bankformat(C));
fprintf('Current Portfolio VaR (diversified):\t $%1s\n',bankformat(VaR_P));
fprintf('Proposed changes in holdings:\n');
for i=1:length(daE)
fprintf(' %s\t $%-s\n',pcE{i},bankformat(daE(i)) );
end
fprintf('New Portfolio VaR (diversified):\t $%1s\n',bankformat(VaR_PaE));
fprintf('Incremental VaR:\t\t\t $%1s\n',bankformat(iVaRE));
generating new risk estimations:
Current exposure: $2,999,997.00 Current Portfolio VaR (diversified): $33,027.94 Proposed changes in holdings: AAPL $0.00 DIS $0.00 IBM $0.00 JNJ $0.00 KO $0.00 NKE $0.00 TXN $0.00 MA $50,000.00 V $50,000.00 New Portfolio VaR (diversified): $33,715.09 Incremental VaR: $687.15
From this example we can see that extending our portfolio with two assets would lead us to the increase of 1-day portfolio VaR by approximately $\$$680. A digestable amount.
Stay tuned as in Part 4 we go extreme with Extreme VaR.
Endnote
When you pilot an aircraft from Amsterdam to Tokyo, the path is usually over Norway, North Pole, and eastern Russia. If one of the engines goes down, you switch it off and ground the plane at the nearest airport. The risk in a commercial aviation is too high to continue the flight.
As a portfolio manager you deal with similar situations but you do not necessarily close your portfolio at once. Instead, you reduce the risk to troublesome assets. You can still fly in the markets but with reduced thrust and hopes for big gains. But don’t worry! Every plane lands with its nose up towards the sky. You can be losing but stay positive till the very end. You never know when cross-winds will change your situation. You may be in the game again before the bell will close the session.
