/*----------------------------------------------------------------------------

The FORECASTING TENSOR yxf: Example Calculation for AR(3) - move back the diagonal
and up after reaching column 0.

yxf  |  h=0   h=1   h=2
-----------------------
t=0 | [1.0]  0.0   0.0
    |      \     \
t=1 | [2.0]  1.1   0.0
    |      \     \
t=2 |  3.0  [1.9]  0.9
    |      \     \
t=3 |  4.0   3.1  [2.1]
    |      \     \
t=4 |  5.0   3.9   2.9
    |      \     \
t=4 |  6.0   5.1   2.9

yf(3, 2, 0) = a + rdot = a + b(0) * yxf(2, 1, 0) + b(1) * yxf(1, 0, 0) + b(2) * yxf(0, 0, 0)

Each sheet represents a variable (n1 endogenous (y) and n2 exogenous (x)),
e.g., endogenous: new_deaths, new_deaths^2
      exogenous:  icu_patients, hosp_patients, new_tests, people_vaccinated, people_vaccinated^2

TensorD: time x horizon x variable
         170    4         7

Model: max lags per variable; selected lags per variable

Note: 'a' is the constant term and rdot multiplies the parameter vector 'b' times
elements in a diagonal in reverse.  Also, the upper right triangle is unknowable
unless back-casting is used.

Column h = 0: zeroth horizon forecasts are the actual (e.g., today's known) values in the time series
Column h = 1: horizon one forecasts are the one-step ahead (e.g., tomorrow's) forecasts
Column h = 2: horizon two forecasts are the two-steps ahead (e.g., day after tomorrow's) forecasts

Row time t = 3: yxf(3, 0, 0) = 4.0 = the actual value for day 3,
                yxf(3, 1, 0) = 3.1 = the one-step ahead forecast for day 3, made yesterday
                yxf(3, 2, 0) = 2.1 = the two-steps ahead forecast for day 3, made two days ago

----------------------------------------------------------------------------*/