My mom makes great spaghetti sauce from scratch. She uses fresh Roma tomatoes, imported virgin olive oil, fresh garlic, parsley, and oregano - when she has the time. When she doesn't have time to make from scratch, she does what most mothers do: she buys sauce from the store.

Howard Moskowitz is a Harvard-educated scientist who made food-testing history twenty years ago for his then radical opinions about how to sell spaghetti sauce. In 1986 he was hired by the Campbell Soup company to develop new recipes for their Prego spaghetti sauce. Moskowitz developed empirical models that analyzed such inputs as spiciness, sweetness, tartness, thickness, and "mouth feel" to determine the best spaghetti sauce ingredients. He found that spaghetti sauce consumers fall into three categories: those that like plain sauce, those that like spicy sauce, and those that like chunky sauce. Prego brought chunky sauce to the marketplace, an innovation back in the mid-eighties, and promptly took huge market share from Ragu. And Moskowitz made food history for his research.

Moskowitz helped set the stage for the spaghetti sauce wars because he used empirical data and statistical prediction - better known to readers of this blog as "naked strategies" - to predict what type of consumer would like which sauce best.

Back to my mom. . . While she sometimes buys Prego, she feels compelled to alter it before serving. She'll add olive oil, grated cheese, sugar, or whatever "is needed" to improve the store-bought stuff. When she does this, she makes the sauce "hers," although she never feels it's as good as it would be if she had made it from scratch. But, by adding her own ingredients, she feels that at least she's making it better.

. . .I doubt if adding any of the extra ingredients actually makes the sauce score any higher on Mr. Moskowitz's statistical models though. After all, these sauces are optimized to hit the target palates. And here's my point: many people take well-developed statistical prediction models - developed for security selection and that have been carefully constructed - and make them theirs by second-guessing the results. In doing so they hardly ever improve the results.
Leading investment strategist Richard Cripps, who has spend years developing and testing a statistical model for security selection, tells me every time he shows the model to a security analyst, they say: "Wow, this is great. Now this is how by interjecting my judgment I can make it better."

Seems like Italian moms and security analysts need more convincing about why "naked strategies" are just fine for security selection or spaghetti sauce.