### Highlights fromSimulate a Hawkes process

from Simulate a Hawkes process by Dimitri Shvorob
(and visualize it)

HAWKESDEMO: Simulate and visualize a Hawkes process

# HAWKESDEMO: Simulate and visualize a Hawkes process

(one of constant-unconditional-intensity, first-order, exponential-decay type)

## Univariate process 1

Self-damping: alpha < 0

par1.mu    =  0.5;
par1.alpha = -0.1;
par1.beta  =  1;


## Univariate process 2

Self-exciting: alpha > 0

par2.mu    = 0.5;
par2.alpha = 0.1;
par2.beta  = 1;


## Simulated histories of univariate processes

t = 10;

h1 = simhawkes1(t,par1);
h2 = simhawkes1(t,par2);


## Intensity-and-event-occurrence plots

showhawkesm(1,t,h1,par1)
suptitle(['Univariate Hawkes process, ' ...
'\it\lambda(t) = \mu +  \Sigma_{\tau_{i}<t} ' ...
'\alpha exp\rm[\it-\beta(t-\tau_{i})\rm]'])

showhawkesm(1,t,h2,par2)
suptitle(['Univariate Hawkes process, ' ...
'\it\lambda(t) = \mu +  \Sigma_{\tau_{i}<t} ' ...
'\alpha exp\rm[\it-\beta(t-\tau_{i})\rm]'])


## Simulated history of a bivariate process

Alas, a multivariate simulator is not yet available. (Can you find a clear recipe? Please send me the reference). We put together two previously generated univariate histories, creating a path of a 'mutually independent' bivariate process.

H = {cell2mat(h1) cell2mat(h2)};


## Bivariate process

.. but plot intensities for a 'mutually dependent' one, to check that inthawkesm and showhawkesm routines work correctly. Note that one-way dependence is engineered: series-1 events affect series-2 intensity, but not vice versa.

par.mu    = [par1.mu    par2.mu];
par.alpha = [par1.alpha 0.0; ...
0.5        par2.alpha];
par.beta  = [par1.beta  0.5; ...
0.5        par2.beta];


## Intensity-and-event-occurrence plots

showhawkesm(1:2,t,H,par)
suptitle(['Multivariate Hawkes process, ' ...
'\it\lambda_{i}(t) = \mu_{i} + \Sigma_{j} ' ...
'\Sigma_{\tau_{ji}<t} \alpha_{ij}\rm ' ...
'\itexp\rm[\it-\beta_{ij}(t-\tau_{ji})\rm]'])


Example: none (Oh, the FEX code metrics..)