What if policymakers could predict the tipping point of a crisis and act before violence erupts?
In a world increasingly shaped by AI-driven forecasts, one crucial area lags behind: conflict prevention. In a new working paper from the Barcelona School of Economics, “If You Only Have a Hammer: Optimal Dynamic Prevention Policy,” Hannes Mueller, Christopher Rauh, Alessandro Ruggieri, and Ben Seimon explore how integrating machine learning predictions into policymaking can drastically improve prevention strategies and reduce long-term economic and humanitarian costs.
Why Prevention Often Is Overlooked
Public policies usually react to crises rather than preventing them, despite the well-documented costs of armed conflict. Using a dynamic decision model, the authors show that policymakers face a dilemma: intervene early at a relatively low cost or wait until violence escalates—when interventions become expensive and politically challenging.
At the heart of their approach is a Markov model that maps out risk stages (see Figure 1) from stable peace (stage 1) to full-scale conflict (stage 12). Their findings reveal a paradox: preventive action increases the likelihood of successful de-escalation later on, but paradoxically, de-escalation efforts can reduce incentives for prevention.
Thinking about dynamics is important. The probability of a conflict can be low, but the costs tend to be very high. This is not just because of the immediate costs in Panel A of Figure 2 but because of the conflict trap: one conflict typically does not come alone. If one experiences conflict once, the likelihood that one experiences conflict in the near future again (and again) is considerable. Therefore, the net present value of future (expected) cost can be high even during relatively low risk situations as can be seen in Panel B of Figure 2.
NOTES: Both panels show the average damages Ds in a given period for panel (A) countries with below
median GDP per capita and (B) above median GDP per capita. Note that in the partial information
case shown here, the perceived future damages in stages 1-5 are always equal since the policymaker
cannot distinguish these stages. Therefore, we modify the static damage vector for stages 1-5 to be a
weighted average of the static damages in the full information case, according to the in-group ergodic
distribution of the full information transition matrix in the absence of interventions.
The Trillion Dollar Question: What’s the Cost of Late Action?
Through simulation, the study estimates that failing to act on early warning signals carries an information rent—potential economic gains lost due to lack of forecasting—equivalent to 60% of a country’s GDP.
In other words, a policymaker who lacks clear risk differentiation is flying blind, missing opportunities to prevent crises before they spiral out of control.
The study also finds that preventive interventions yield the highest return on investment just before conflict risk spikes, particularly in post-conflict or high-tension situations as can be seen in Figure 3. The benefit-cost ratio (BCR) of intervening in these moments can increase 15 times the investment. Prevention and de-escalation interact. If de-escalation policies are overly relied upon, the BCR for prevention drops by nearly 33%, reinforcing the need for a balanced strategy.
NOTES: Figure shows reports benefit cost ratios (BCRs) from 10,000 simulations across all countries
by stage calculated from equation (4.2). For each simulation we draw a transition matrix and damage
vector, calculate the BCR for each country, and average across countries. Point estimates are the mean
BCR score. Bars represent the 2.5th and 97.5th percentile of the 10,000 means. Calculations include
GDP damages.
Lessons for the Real World
From climate change mitigation to pandemic preparedness, policymakers have learned that early action pays off. The same logic should apply to conflict prevention. Yet, today’s policies still mostly react to violence rather than anticipate it.
By leveraging predictive models, governments and international organizations can turn crisis response into crisis prevention—saving lives, stabilizing economies, and avoiding the spiral of humanitarian disasters.
With geopolitical tensions rising worldwide, from Ukraine to Sudan, the question isn’t whether we can afford to invest in prevention—it’s whether we can afford not to.
