Winding annual-report narratives from loss-making firms track later earnings upturns

Each MD&A section was cut into chunks of roughly 250 words. Each chunk became a point whose position is the average of its word vectors, so chunks about similar things sit close together ^[5][7]. Keep the points in reading order, trace the path between them, and divide its length by the shortest path that would have visited every point. That ratio is circuitousness ^[6][7]. The chunking follows a method from marketing research by Olivier Toubia of Columbia and Jonah Berger and Jehoshua Eliashberg of the University of Pennsylvania ^[5].

Guest, the Cornell co-author ^[3], traces the idea to the "spiral curriculum," in which "a teacher should start simple, introduce a concept, and then come back to it at different times, contextualized maybe at a higher difficulty level as they go," he said ^[15].

The sample is more than 13,000 MD&As filed with the SEC from 1997 through 2019 ^[4]. That is 23 years, at an average of upward of 565 filings a year ^[16][17]. The authors focused on firms with negative earnings and compared circuitousness with the Fog index, document length and repetition ^[8]. Those are the controls a skeptic asks for first, since a longer or more repetitive document has more room to wander. The published summary does not report effect sizes, how the new measure fared against those three indices, or how many of the filings came from loss-making firms.

The authors' own hypothesis was that circuitousness is more common when a firm is struggling but close to recovery, because it has a lot to explain ^[10]. The result fits: the association shows up for loss firms that later turned around and is absent for those heading into a downturn ^[1]. On that account, managers who can see a recovery coming have more threads to connect, and the prose follows the prospects. The evidence does not show that writing in loops helps a firm with no recovery to describe.

Circuitous filings also drew more EDGAR downloads ^[9]. The thing this doesn't tell you is whether the people downloading those filings understood the firms better, or traded on what they read.

Guest makes the payoff conditional. "CEOs shouldn't necessarily try to be as brief as possible," he said ^[11]. "Talking about topics such as revenues in multiple contexts ... might actually be useful," he said, "if there is an investor who has the time and energy and sophistication to sit down and process it all" ^[12]. That processing costs cognitive effort or computing resources, Guest said ^[10].

He also separates circuitousness from obfuscation ^[13], while granting that managers have reasons to blur even good news. "If they release all the good news right away," he said, "they might attract the attention of competitors, who might say, 'Hey, why don't we copy what they're doing?'" ^[14]

I think the paper supports a narrower claim than a general case against brevity. In loss-making firms' MD&As, a narrative that keeps returning to its topics is associated with a recovery that followed and with more readers ^[1][9]. That makes it a plausible screening feature for analysts able to compute it from public filings ^[4].