The historical correlations between 7-day moving averages of covid deaths and hospitalizations and ICU beds are assumed to hold. What does this mean for future trends in coronavirus-related deaths?
figure 1: Covid deaths, 7-day moving average (black) and 7-day moving average (red) of predicted lagged hospitalizations and ICU deaths based on log-level OLS regression, 95% confidence intervals (see below). Source: Our Data World, accessed 16 January 2022, author’s calculations.
Point estimate for 1/28 of 3176 deaths (7-day moving average).
I use a sample from August 5, 2020 to January 15, 2022 to estimate:
deceasedTon = 357.08 + 0.816lhospt-21 + 0.300 Facet-14 – 0.054trend + 0.000002trend2
Adj R2 = 0.94, SER = 0.165, DW = 0.09, N = 529, bold indicates significance at 5% msl, using HAC robust standard errors.
This ad hoc specification is based on the following observations:
figure 2: Covid deaths, 7-day moving average (black, left scale), hospitalization time lag 21 days (red, right scale). Source: Our Data World, accessed 16 January 2022.
image 3: Covid deaths, 7-day moving average (black, left scale), ICU use lag of 14 days (red, right scale). Source: Our Data World, accessed 16 January 2022.
Since the in-log-levels norm exhibits a high degree of serial correlation (possibly spurious regression), I estimate first-order differences.
ΔdeceasedTon = 0.00045 + 0.276Δlhospt-21 + 0.270Δ Facet-14
Adj R2 = 0.07, SER = 0.048, DW = 1.69, N = 529, bold indicates significance at 5% msl, using HAC robust standard errors.
(Coefficients for linear time trends are not statistically significant.)
The norm yielded a lower forecast of 2342 at 1/28, which would be close to the fall 2021 peak (9/30).
Figure 4: Covid deaths, 7-day moving average (black) and log-based prediction OLS regression lagged 7-day moving average of hospitalizations and ICU deaths (red), and log-based first difference (green) (see above) . Source: Our Data World, accessed 16 January 2022, author’s calculations.
