This project has code written in matlab, python and q/KDB+ code. Code mirrors each other

My contribution was to convert matlab file over to q/KDB+, for Kalman gain portion.

Code refers to equations from: Book: Ernest P Chan "Algorithmic Trading - Strategies and their Rational" (2013)

Matlab file implements examples of computing beta between two ETFs, using kalman gain.

speed: python is slowest. Then matlab. Fastest is q/kdb+.

to time in kdb+/q/KDB

q)\t \l beta.kalman.q

• Note: xAT is Transform of xA
• Note: This script was run in kdb+ 64x, cmd line: q q.k -s 1 -p 5010

======Code flow analysis

+  R  Estimated measurement error covariance. 
+  Q  Estimated process error covariance, measurement variance prediction
+  K  Kalman Gain
+  P  State Variance
+  e  measurement prediction error ( This is 'M' in other literatures. )
+  yhat measurement prediction

• ========================Init Data ================================

• ===measurement/model equations===**********=====system/process equations============

•   									| Vw:Vw*(delta%(1-delta))
    									| Ve:0.001;
    | xA:EWA cls px								
    | yC:EWC cls px
    | beta[;0]:0f;
  

• ------------Start of LOOP through each Data Point of EWA/EWC Pair for its beta --------

• ===measurement/model equations===**********=====system/process equations============

• =======transform of data=====*************======Variance/Covariance Adj======

•   								  | R:P+Vw
    | beta[;t]:beta[;t-1] (if not initial loop)
    | yhat,:sum xA[t;]*beta[;t]
    								  | Q,:(sumMV[R;xA[t;]]) + Ve	       Q=xA.R.xAT + Ve	
    | e,:yC[t]-yhat[t]
    								  | K:mmu[R;vvmu[xA[t;];(1%Q[t])]]   K=R.xAT.1/Q
    		 _______________<_Adj Var__________________________|				
    	            |
    | beta[;t]:beta[;t]+K*\:e[t]
    								  | P:R-vvmu[mmu[xA[t;];R];K]  		P=R-K.xA.R
  

• -----------------------------------End of LOOP----------------------------------