Organization

• Find the one central, new contribution of your paper and write it down in a single concrete paragraph: state the result itself, not that you found interesting results.
• Readers skim. Put the main result first and fill in detail afterwards, as a newspaper does, instead of building to a punchline in the last table.
• Leave out the story of how you got there and the things that did not work.

Abstract and introduction

• Keep the abstract to 100–150 words. Say concretely what you find, and leave other literature out of it.
• Open the introduction with your contribution, not with philosophy, the literature, or policy motivation.
• Give the fact behind each claim, such as the regression and the number, not just the conclusion it supports.
• About three pages is enough for an introduction. A roadmap paragraph is optional.

Literature review

• Explain your own paper before comparing it with others: readers cannot judge the difference until they understand your contribution.
• Keep the review short and set apart. Focus on the two or three closest papers and on giving credit where it is due.
• Be generous. Your paper does not need everyone else to be wrong in order to be interesting.

Body of the paper

• Nothing should come before the main result unless the reader needs it to understand that result.
• In an empirical paper, include only the theory the empirics need, and write down the specific model you take to the data rather than a general one.
• Lead with the main result, follow it with the graphs and tables that show why it holds, then a limited set of robustness checks.
• Keep the conclusion short: no full restatement, no research agenda, no speculation.
• Move extra robustness checks, extensions and the general model to an appendix, online if need be.

Writing

• Keep it short. Every sentence should earn its place, and a finished paper should rarely need more than about 40 pages.
• Say each thing once. Phrases like “in other words” usually signal a repeat.
• At every level, first say what you do, then why, then how it compares with the alternatives.
• Frequent pointers forward and back, such as “as we will see” or “recall that”, suggest the material is in the wrong order.
• Document your work so that another student could reproduce every number from the paper and its appendix.
• Simpler is better. Less math and simpler methods make a paper more convincing, not less impressive.
• Keep track of what the reader already knows. Most students assume far too much.

Footnotes, tables and figures

• An aside that matters belongs in the main text, and one that does not should go. Footnotes are for material most readers can safely skip.
• Give every table and figure a caption a skimming reader can understand on its own. For a regression, include the equation and the dependent variable.
• Discuss every number you show. If it is not worth a sentence, drop it from the table.
• Report two or three significant digits, in units that give readable numbers.
• Label axes, define every symbol in the caption, and choose line styles that still show up when printed.

Style

• Use the active voice and take responsibility for your claims. “I” is fine on a sole-authored paper.
• Prefer the present tense, and keep tenses consistent within a paragraph.
• Write plain subject–verb–object sentences with few clauses, short words and little jargon.
• Do not praise your own results with adjectives; let readers judge them.
• Cut empty openers such as “it should be noted that” and just make the point.
• Follow “this” with a noun, so the reader knows what it refers to.
• Give Greek letters names as well as definitions, and remind readers what they stand for.
• Describe a model’s structure directly; keep “assume” for claims about the real world.
• Spell out authors’ names, skip the opening quotation, and use italics sparingly.

Empirical work

• Identification comes first. Explain what moves your right-hand-side variable, what makes up the error term, and why the two are unrelated.
• Explain why your instruments move the regressor but not the error, and know when a variable belongs as an instrument rather than a control.
• For every estimate, say which variation in the data drives it, for example within firms over time or across firms.
• Ask whose behavior you are modeling, so you do not mistake a supply curve for a demand curve, and rule out reverse causality.
• Use few controls. Do not control for the channels through which your variable works, and be suspicious of a very high R².
• Show the stylized facts behind your estimates, not just test statistics.
• Report economic magnitudes as well as statistical significance, and a standard error for every important number.

Seminars

• Skip the motivation, literature review and preview of results; start with the main result.
• Put equations, tables and graphs on slides rather than your script, and leave each slide up long enough to absorb.
• Listen to the whole question and pause before answering. Keep paper at hand to note what to change.
• Err toward simplicity: audiences absorb far less in one sitting than presenters expect. Speak slowly and clearly, and finishing early is fine.

Why it matters

• Economics papers are essays, and good economists spend much of each project on the writing. Cochrane points to Zinsser’s On Writing Well and McCloskey’s Rhetoric of Economics for more.