Slotting allowances: a time series analysis of aggregate effects over three decades

The BATF likened slotting to a consignment sale, noting “that the practical effect of slotting allowances is to refund, in whole or in part, the purchase price of a product that has not been sold, in proportion to the period of time that it remains unsold” (Federal Register 1995).

Data on new products and selling area for “all other retailers” (to form a parallel control group) are not available, consequently the hypotheses here are framed only in the context of differences across t₁ and t₂. Further, data for innovative new products are available only for t₂ starting in 1985.

Goodale stated in the Federal Register (1995, p. 6) that slotting allowances accounted “for perhaps more than 10% of after-tax profits” (see also footnote 5 in the article). The Wall Street Journal reported that “slotting fees can represent 20% to 40% of the profits of some chains” (Jenkins 1999). Allain and Chambolle (2005) show that on average 88% of the margins earned by French supermarkets on grocery products in 1999 were of the “hidden kind” (slotting fees and conditional rebates) compared to conventional rebates and margins. Retailers are becoming more open about the relationship between promotional fees, bottom lines and retail prices. Donald Sussman (Executive VP Purchasing, Ahold USA) says: “Every dollar of slotting we generate is a dollar we don’t have to generate” from sales. If slotting fees were not there, “the pressure on prices would rise” (quoted in EITF 00-25, WGR #5, April 4, 2001, p. 18). Mark Polsky (Senior VP, Magruder’s, MD) notes: slotting “goes down to the bottom line”; retailers need the money to manage their bottom lines, and without it “you can either cut your help or raise prices” (quoted in the Washington Post 2004).

The analytical modeling literature is split in its derived theoretical outcomes. Some models show profit shifting to the party with the bargaining power (Chu 1992; Messinger and Chu 1995; Marx and Shaffer 2004). Others show that slotting fees do not grant retailers any excess profit; rather, a fee is offered by the manufacturer only when the retailer faces a high opportunity cost to stock a product (Lariviere and Padmanabhan 1997). Kuksov and Pazgal (2007) show that competition among retailers drives slotting and that in the equilibrium, retailer profits may be negatively correlated with slotting fees (see also Desai 2000).

It is note worthy that in retailers’ accounts COGS has always been recorded net of trade discounts. A study of financial restatements by resellers in response to the FASB (2002) edict shows that practically all of them report they follow this practice and therefore did not retrospectively restate income.

This includes all firms in the following SIC retail groups (after removing internet only and non-U.S. firms): 5,200 Building Hardware Garden, 5,211 Lumber & Building, 5,311 Department Stores, 5,331 Variety Stores, 5,399 Misc. General Merchandise, 5,531 Auto & Home Supplies, 5,600 Apparel & Accessories, 5,621 Women’s Clothing, 5,651 Family Clothing, 5,661 Shoes, 5,700 Furniture & Equip., 5,712 Furniture, 5,731 Radio, TV & Electronics, 5,734 Computer & Software, 5,735 Records & Tapes, 5,912 Drug Stores, 5,944 Jewelry, 5,945 Hobby & Toys, 5,990 Misc. Retail Stores (e.g., Office Depot, PetSmart).

In COMPUSTAT this variable is listed as OIBDA—Operating Income before Depreciation and Amortization. This measure is gaining ground as companies move away from using EBITDA—Earnings before Interest, Taxes, Depreciation and Amortization. The measures are alike except EBITDA includes non-operating income whereas OIBDA does not.

Perron considers three hypothesized process: A–where the interruption event causes a shift in the level of the series, B—the event causes a change in the trend or rate of growth and C—where both level and rate of growth are expected to change. Slotting fees grew gradually over time and are more likely to have resulted in a change in the rate of growth rather than a level shift. Nevertheless the series are first modeled with μ, β, θ and γ, and if one or more are insignificant they are dropped sequentially and the model re-estimated. Equations 2B and 2C and corresponding c.v. in Perron’s Tables V.A and VI.A are found to be applicable in the analyses of unit roots in this study.

Equation 2C gives the following results: ρ = .953, T(ρ-1) = −18.08 and ρ = .952, T(ρ-1) = −18.59 for the CPI-f and CPI-nd respectively. Both fall within the critical value −32.47 for λ = 0.3 (Perron Table VI.A). For the CPI-f and CPI-ab Eq. 2B applies and gives ρ = .968, T(ρ-1) = −10.4 and ρ = .976, T(ρ-1) = −6.5 respectively, within the critical value -28.68, λ = 0.7 (Perron Table V.A).

The best fitting model for ΔCPI-f is k = 6, but all χ² to lag 24 are significant, and the LM test rejects the null (Obs*R² = 95.24, df = 24, p = .0000). For ΔCPI-nd the best fit is with k = 5; but six χ ² are significant although the LM test fails to reject the null χ² Obs*R² = 19.84, df = 24, p = .706. In the ΔCPI-f v. ΔCPI-ab analysis, the best fitting model for ΔCPI-f is k = 0 but 15 significant χ² remain to lag 24 and the LM test rejects the null of no serial correlation (Obs*R² = 61.11, df = 24, p = .0000). For ΔCPI-ab, k = 1 all χ² up to lag 8 are significant even though the LM test is insignificant (Obs*R² = 32.37, df = 24, p = .118).

This perspective is thanks to a reviewer.

Quasi rent is defined as the income earned above the opportunity cost of a sunk investment. If the sunk investment is highly specialized (e.g., a bridge, a patent, or a retail store in a particular location) the next best use may not offer much return. Public policy is sometimes used to ensure that such types of sunk investments earn large quasi rents so that new investments continue to be made in existing and new products of the kind.

It is customary to measure industrial concentration as the market share of the four largest firms. However, a 20-firm ratio is considered more descriptive of concentration in retailing because retail chains continue to be regionally dominant rather than nationally.

For example: “CG companies are struggling with the prospects of having line items and entire brands (in some cases entire companies) delisted from the shelf lineup of large retailers such as Safeway, Kroger, Wal-Mart, Walgreen, 7-Eleven, Target and others. These retailers have been the engine of growth for CG companies for over 20 years. The retailers, following each other and now being followed by mid-tier players are all pursuing at least a two prong strategy to bolster profits and build a stronger bond with their target customer segment. Prong one is the systematic reduction of entire national brands or sizes/flavors of national brands in order to reduce stocking and inventory costs. Prong two is the introduction of a stronger lineup of Private Label items” (Spindler 2010).

U.S.C. Section 13(C): “Payment or acceptance of commission, brokerage or other compensation. It shall be unlawful for any person engaged in commerce, in the course of such commerce, to pay or grant, or to receive or accept, anything of value as a commission, brokerage, or other compensation, or any allowance or discount in lieu thereof, except for services rendered in connection with the sale or purchase of goods, wares, or merchandise, either to the other party to such transaction or to an agent, representative, or other intermediary therein where such intermediary is acting in fact for or in behalf, or is subject to the direct or indirect control, of any party to such transaction other than the person by whom such compensation is so granted or paid.”

A DSP means ρ = 1, and β = γ = θ = 0. But there are DGP where ρ = 1 but β, γ, θ ≠ 0. Further, differencing does not always result in an error stationary model. Differencing can result in a series characterized by a moving average even when the original series do not have a cycle (Maddala and Kim 1998, p. 13). A TSP modeled in first differences leads to overdifferencing; conversely a DSP estimated in levels leads to underdifferencing (Maddala and Kim 1998, p. 88). Maddala and Kim state it is “more important to take the serial correlation structure in both the models into account…”

This is similar to Ben-David and Papell (1995, p. 464) who compare the pre break to the post break series after eliminating an endogenously determined “transition” period.