Economics Prompts

Using Economics Prompts to Cut Through the Noise in Analysis

If you've ever tried to get a useful output from an LLM on a macroeconomic question, you know most models just give you generic textbook answers. They hedge. They summarize Wikipedia. They don't actually work through the mechanics of what you're asking. That's where specialized Economics Prompts come in, and that's also why the free ones floating around Reddit and Discord threads are usually useless. I started building my own prompt templates about two years ago because I was tired of feeding a model a GDP growth question and getting back a paragraph that could have been written by any intro student. The difference between a bad output and a usable one usually comes down to three things: you force the model to show its work in economic terms, you specify the framework you want it to reason through, and you give it boundary conditions instead of open-ended freedom.

Why Most Economics Prompts Fail on First Try

Here's the thing nobody admits — a raw prompt like "Analyze the effect of rising interest rates on inflation" will almost always produce a vague, balanced, deeply unhelpful response. The model defaults to hedging because that's what it was trained to do. It's not broken. It's doing exactly what it was optimized for. The fix is specificity. Start by telling the model which framework to apply. DSGE? IS-LM? Supply and demand with price stickiness? Agent-based? Pick one and make it commit. When you constrain the reasoning structure, the output jumps from generic to actually analytical. I've seen this cut my revision time from three reads down to one. You also need to specify the jurisdiction and the time horizon. An interest rate analysis for the US Federal Reserve in 2023 is completely different from the same question applied to the ECB in 2011 or emerging markets in a dollar crisis. The model won't assume any of this unless you tell it. I lost about a week last year going back and forth on a model output that turned out to be describing the Bank of Japan's yield curve control framework when I'd only asked about "low rate environments." That was entirely on me for not specifying.

Building a Prompt That Actually Works

Let me walk through how I structure an Economics Prompt for something like forecasting or policy analysis, because that's where the most people fall short. First, define the role. Not "you are an economist" — that's pointless. Use "you are a macroeconomist specializing in monetary policy transmission in advanced economies." The narrower the role, the better the output converges on something usable. Second, lay out the data you're working with. Paste the actual numbers if you have them. Fiscal deficit as a percentage of GDP, central bank balance sheet size, inflation expectations from surveys, unemployment rate, whatever's relevant. The model can't hallucinate useful analysis from thin air. I keep a running spreadsheet of current indicators for major economies and paste the relevant rows into my prompts. This alone accounts for probably 60% of the improvement I see.

Third, tell it exactly what form the answer should take. Do you want a structured breakdown? A comparison table? A risk assessment with probability weighting? If you don't specify format, you get prose walls. I usually ask for three sections: baseline scenario, upside case, downside case, each with the key assumptions listed separately. This forces the model to separate its reasoning from its conclusion, which is something most people skip and then wonder why the analysis falls apart under scrutiny. Fourth, include a constraint against generic hedging. Tell it explicitly not to say things like "the impact depends on multiple factors" without then enumerating those factors and assigning relative weight. I know that's blunt but it works because the model will otherwise default to safe, empty language.

A Real Case Where It Got Messy

Last spring I was working through a prompt to analyze the impact of carbon pricing on manufacturing competitiveness in the EU. The initial outputs were surprisingly poor because the model kept conflating the EU Emissions Trading System with a direct carbon tax. Those are fundamentally different mechanisms with different incidence properties, and the model treated them as interchangeable. The workaround was to embed a definitional constraint directly in the prompt. I wrote out the distinction between the two policy instruments in the system instructions and explicitly told the model to flag when a question involved cap-and-trade versus tax structures. After that, the outputs improved dramatically. It's a small thing but it caught an error pattern I hadn't noticed before — the model has a persistent bias toward simplifying distinct policy tools into a single category when they're not clearly separated in the prompt. This matters because the confusion isn't academic. Carbon tax revenue recycling and cap-and-trade auction revenue handling have different effects on firm-level margins and on the political economy of policy adoption. Getting the mechanism wrong in your analysis means getting the prediction wrong too.

What Economics Prompts Can't Do

Let me be clear about the limitations because people oversell this. Economics Prompts will not replace judgment. They will not give you accurate forecasts. They will not resolve disagreements between economic schools of thought — if anything, they tend to flatten those differences into a bland consensus view that sounds reasonable and means nothing. The model's training data cuts off at a fixed point. Any analysis it produces involving data past that cutoff is either fabricated or clearly outdated. I've caught it referencing GDP figures from before the latest quarterly revision more than once. Always verify the underlying numbers independently. A 15-minute fact-check saves you from looking like an amateur. There's also the issue of model drift. The same prompt that worked well in January produced noticeably different quality outputs by March after what appeared to be a model update. I stopped treating any single prompt as permanent. I revisit and adjust my templates every quarter, sometimes more often if a new model version comes out and I notice the output quality shifting.

If you need real-time economic data or highly localized analysis — say, county-level employment trends or a specific sector's earnings call synthesis — you're better off combining the prompt framework with a data API or doing the legwork manually. The prompt is a reasoning scaffold, not a data source.

Getting Started Without Overcomplicating It

You don't need a fancy system to start. Take one question you actually care about — a policy you're following, a market you're watching, a class you're studying — and write a single prompt that includes role definition, current data, required framework, and output format. Run it. Read the output critically. Note where it went soft or vague. Revise the prompt to close those gaps. Repeat. The Economics Prompts you build will get better with each iteration because you're mapping the failure modes yourself. That's the actual value — not the template, but the process of learning where the model slips and how to catch it. I've got maybe two dozen templates now, most of them narrow enough to be useful for one specific type of question. The ones I use every day are the ones I've spent the most time debugging. There's no central repository or official download for these because they're inherently personal. What works for someone analyzing Fed policy won't help you much with trade elasticity questions. The closest thing to a starting point is just writing your first one badly and then fixing it until it stops embarrassing you.