When doing a Transient (.TRAN time domain) simulation, all Spice variants perform an Initial Conditions DC solution to establish where all nodes in the circuit sit at time=0. In the sim above, the stimulus is a square wave, which takes on only two values, 0V or 5V. At Time=0, the input Pulse Voltage source has a value of 0V, so the Initial DC solution causes all nodes in the filter to settle to near ground.
The duty cycle of the Pulse source is 50%, so the average DC level is near 2.5V. Now it takes a 2Hz filter several seconds to settle to the final value when the input steps from 0V to an average value of 2.5V. To avoid showing the startup transient, and waiting for thousands of cycles of the 1000Hz PWM input, waiting for the filter output to settle, knowing that the expected level is tending toward 2.5V, it is easier to just force the internal nodes in the filter to an initial condition of 2.5V before allowing the simulation to run. That way, it is easier to plot and show the ripple component riding on the dc level...