Day Trading & Forex Forum

We see in Figure 3 that both the 2 bar cycle and 3 bar cycle are suppressed.  Since N=4 in this case, the lag of this FIR filter is 1.5 bars.  We can draw the general conclusion that the 2 bar cycle is suppressed only when N is even.  Continuing to increase the order of the FIR filter, using even orders, a 6 bar weighted coefficient filter could be:

C = [1 2 3 3 2 1]/12

The interesting characteristic of this FIR filter is that the 2 bar, 3 bar, and 4 bar cycles are all suppressed, as shown in Figure 4.  The lag penalty for this 6th order filter is 2.5 bars.

So, again, what is the best order of the filter to be used?  I favor the 4th order filter because it only produces 1.5 bars of lag, and avoidance of lag is generally more important to trading that increased smoothing.  The 4th order filter virtually completely removes the two bar and 3 bar variations in the differential closes.  By eliminating these very short term variations in the differential closes, the Smoothed RSI (SRSI) is nearly free of disconcerting wiggles that lead to whipsaw trades.  The SRSI is compared to the standard RSI in Figure 5 and is also compared to the standard RSI that is smoothed by a 4th order FIR filter in Figure 6.  The EasyLanguage code to compute the SRSI is given in Figure 6.  The EasyLanguage code to compute the SRSI is given in Figure 7.

Conclusion

The RSI can be greatly enhanced by smoothing the differential closes before the RSI function is computed.  Not only are the short term variations removed, but the desirable shape of the indicator is enhanced.  In particular, using even ordered and symmetrically weighted FIR filters, the specific short term variations are nearly completely removed.  I hope this trim on an old and trusted indicator will serve you well.

Figure 1.  A two bar Simple Moving Average Removes the Two Bar Cycle Component (Cycle = 2 / (Normalized Frequency)

Figure 2.  A three bar Simple Moving Average Removes the Three Bar Cycle Component (Cycle = 2 / (Normalized Frequency)

Figure 3.  A four bar Weighted FIR Filter Removes Both the Two and Three Bar Cycle Components (Cycle = 2 / (Normalized Frequency)

Figure 4.  A six bar Weighted FIR Filter Removes the Two, Three, and Four Bar Cycle Components (Cycle = 2 / (Normalized Frequency)