Constructing an investor preference index from two capitalization-weighted series

What the preference series is

The investor-preference-index is a constructed spread that compares six-month log changes in a large-cap index and a broader exchange composite, then smooths the residual into a long-horizon preference reading. In this construction, market-breadth is relative participation inferred from two capitalization-weighted indices instead of from an advance-decline count.

The preference series is built from weekly closes of a 500-name large-cap index and a capitalization-weighted composite of all common stocks on the same exchange, so the residual isolates the mid-cap remainder. At year-end 1996 the exchange composite capitalization was about 6.75 trillion and the 500-name index about 5.5 trillion, leaving more than 1 trillion, or about 15 percent, in the mid-cap remainder.

The leadership hypothesis

The construction hypothesis is that mid-cap names lead early in a bull phase and large-cap names lead in the final stage, contrary to the idea that bull markets end in a speculative frenzy. Mid-cap-preference is that early-cycle condition, in which the broader exchange composite outperforms the large-cap index. Blue-chip-preference is the late-cycle condition, in which large-cap names outperform the mid-cap remainder.

From raw residual to published reading

Each index is first converted to a natural-log scale. The raw-ipi is the six-month-change of the broader composite minus the six-month-change of the 500-name index.

The published preference reading is a 54-week sum of the difference between the 15-week and 33-week averages of that residual, then rescaled by adding 1 and multiplying by 100 so the series typically occupies a band near plus or minus 0.1 before rescaling. That step is the trend-filter: a multi-week moving-average difference of the raw preference series, summed over a longer window so only persistent leadership changes register.

Sample history and overlay behavior

The weekly sample runs from the 1966 inception of the exchange composite through 19 September 1997. The timing test begins on 2 August 1968 because the smoother needs about two and a half years of history. The preference overlay is more volatile before 1980 than after, which the construction notes attribute to a smaller listed universe of about 1,200 names in 1968 and to later concentration of interest in the 500-name index.

Visual inspection of the log large-cap series with the preference overlay, later checked in software, associates steep preference downturns with corrections or crashes rather than treating the overlay as a price-pattern top call.

Timing against an untimed baseline

Rank-rotation is the timing procedure that stays invested, exits, or stays out according to the preference reading rather than a single-index price rule. The timing procedure is compared with an untimed 10,000 large-cap buy-and-hold-baseline that reaches a 88,800 gain by mid-September 1997. Under the published parameters the preference system records a 134,500 net gain and more than 50 percent outperformance versus that baseline.

The same procedure is out of the market for seven extended periods in the 1978 to 1987 window, with each exit preceding or limiting a decline. The source states a 20 percent drawdown in a long correction as a floor for what later drawdowns can reasonably be expected to reach.