9th Financial Risks International Forum

Transcription

1 Calvet L., Czellar V.and C. Gouriéroux (2015) Structural Dynamic Analysis of Systematic Risk Duarte D., Lee K. and Scwenkler G. (2015) The Systemic E ects of Benchmarking University of Orléans March 21, 2016

2 Comments on Structural Dynamic Analysis of Systematic Risk Calvet L., Czellar V.and C. Gouriéroux (2015)

3 Summary This paper introduces an original structural dynamic factor model (SDFM) for stock returns Nonlinear impact of the common factor and stochastic volatilities on returns, that depends on the distance-to-default. Estimation: Indirect Inference (Gouriéroux, Monfort and Renault, 1993) Filtering: Approximate Bayesian Computation algorithm (Calvet and Czellar, 2011)

4 Fact The structural dynamic factor model (SDFM) is a (structural) nonlinear state space model Engle, R., and E. Siriwardane (2015). Structural GARCH: The Volatility-Leverage Connection. Harvard Business School Working Paper

5

6

7 from a Dynamic Factor Model (DFM)... Y it = α i + β i F t + σ it ɛ it F t = γf t 1 + η t u t η 2 t ARG δ, ρ, (1 ρ) 1 γ2 /δ σ 2 it ARG e δ, eρ, ec ARG: AutoRegressive Gamma (ARG): exact time discretized Cox-Ingersoll-Ross process State variables: F t : common unobservable linear factor η t : factor stochastic volatility σ it : rm-speci c stochastic volatilities (i = 1,.., n)

8 to a Structural Dynamic Factor Model (SDFM)... a it = α i + β i F t + σ it ɛ it F t = γf t 1 + η t u t η 2 t ARG δ, ρ, (1 ρ) 1 γ2 /δ σ 2 it ARG e δ, eρ, ec Y it = h (log (A it /L 1)) h (log (A i,t 1 /L 1)) a it = log (A it /L 1) where A i,0 /L is xed and the asset value A it and the liabilities L (?) are unobservable.

9 This model is structural since it is based on an economic de nition of the default from the balance sheet information (Value-of-the Firm model). When the asset/liability ratio is large ( rm is far from the default): Y it ' a it But, in the general case, the stock returns depend not only on the changes a it, but also on the leverage A i,t 1 /L 1 The leverage e ect increases with the proximity to default, as measured by the excess asset/liability ratio.

10 Comment 1 In the paper, the authors derive three risk measures: 1 the implied nancial leverage AL it = ba it /L 1 2 the distance-to-default DD it = bh i log(ba it /L 1) log(ba it /L 1) 3 a measure of the magnitude of speculation ( DD) 2 Why not consider the SRISK index as in Engle and Siriwardane (2015)?

11 De nition (SRISK, Brownlees and Engle, 2015) The SRISK corresponds to the expected capital shortfall of a given rm i, conditional on a crisis a ecting the whole nancial system. SRISK it = W it (k LVG it (1 k) MES it 1) MES it (α) = E(Y it jy mt VaR mt (α); Ω t 1 ) with D it the book value of debt, W it the market value of equity and LVG it = D it / (D it + W it ).

12 Toward a structural-based SRISK

13 Source: Engle and Siriwardane (2015), page 30

14 Comment 2: Indirect inference and validation of the SRISK.. It could be interesting to compare 1 the structural-based SRISK and the "naive" SRISK for di erent rms (time-series analysis), 2 the systemic risk rankings implied by the two SRISKs at a given date (Spearman rank correlation). "Structural" validation test of the SRISK 1 Include SRISK-based statistics (rank correlation, etc.) in the auxiliary model (indirect inference) 2 Compare the estimated parameters with and without these additional statistics

15 Comment 3: The sample includes only 10 nancial institutions. Is it possible to consider a sample with a large n dimension? Comment 4: In the SDFM, the leverage has an impact on the return s volatility, as soon as the rm is close to the default => kind of switching regime model. What is the estimated e ect of the leverage on the volatility for the considered nancial institutions?

16 Systemic E ect of Benchmarking Comments on The Systemic E ects of Benchmarking Duarte D., Lee K. and Scwenkler G. (2015)

17 Systemic E ects of Benchmarking Summary Original theoretical model designed to study the systemic e ects of benchmarking Competitive pressure to beat a benchmark may induce institutional trading behavior ( re sales) that exposes retail investors to tail risk The model is (relatively) simple and elegant, and it can be solved in semi-closed form. So, the model allows to run: 1 a comparative analysis with di erent benchmarking incentives 2 a survival analysis and a welfare analysis.

18 Systemic E ects of Benchmarking Model Pure-exchange economy with two types of investors: retail and institutional Two assets: a benchmark and a non-benchmark stock Retail investors are mean-variance investors The utility of institutional investors increases in the states in which the benchmark stock outperforms. Main result If bad news about the benchmark arrives (jump), the institutional investors iniate re sales. These re sales increase the tail risk exposure of the retail investors.

19 Systemic E ects of Benchmarking Comment 1: The authors assume that the two sources of risk (for the dividends D 1t and D 2t ) are independent. Under this assumption, the systemic e ects of the benchmarking are maximum. If the systemic e ects decrease with the correlation of the two assets, it could be interesting to discuss this point.

20 Systemic E ects of Benchmarking Comment 2: The authors measure the systemic e ects of the benchmarking through the VaR of the nancial system. If we decompose the aggregate VaR between the institutional and retail investors, who have the largest contribution? Assumption: the retail investors have the largest risk contribution when the benchmark importance parameters are large and s 0 is small This is puzzle for the regulator since the largest risk contributors are not at the source of the systemic risk What are the best regulatory tools for the asset management industry within this framework?