Central Banking in Practice

Level: Graduate | Field: Monetary Economics

1. Independence and mandates

The time-inconsistency problem (Lecture 6) motivates delegating monetary policy to an independent central bank with a clear mandate (Rogoff, 1985). Cross-country evidence shows a negative correlation between legal independence and inflation in advanced economies (Alesina and Summers, 1993), weaker evidence for emerging markets, where de facto independence (for example, governor turnover) matters more (Cukierman, Webb and Neyapti, 1992). Garriga (2016) and Romelli (2022) provide updated indices. Mandates differ: the Fed has a dual mandate of maximum employment and stable prices; the ECB a primary price-stability mandate; the Bank of Korea a price-stability mandate with attention to financial stability; Bangladesh Bank broader development objectives.

2. Inflation targeting: does it work?

Inflation targeters have had lower and more stable inflation since the 1990s. But so did non-targeters. Ball and Sheridan (2005) show that once one controls for initial inflation (regression to the mean), the advantage of inflation targeting in OECD countries largely disappears. Studies of emerging markets find larger effects (Gonçalves and Salles, 2008), but adoption is endogenous: countries tend to adopt inflation targeting at the end of a reform process — after granting independence, reducing fiscal dominance and abandoning pegs — so pre-adoption trends must be modelled. Modern staggered difference-in-differences and event-study estimators (Callaway and Sant'Anna, 2021; Sun and Abraham, 2021) address the bias of two-way fixed effects with heterogeneous adoption dates. Exercise 15.1 reproduces the Ball–Sheridan logic with World Bank data.

3. Inflation expectations and communication

Because the New Keynesian model is forward-looking, expectations are the main transmission channel, and communication is a policy instrument. Central banks publish forecasts, policy-rate projections (the Fed's dot plot, the Riksbank's and Norges Bank's rate paths), minutes and press conferences. Long-term expectations in surveys and break-even inflation stayed relatively anchored in 2021–23, which helped bring inflation down without a deep recession. Households' and firms' expectations, however, are dispersed, respond to salient prices such as food and gasoline, and are often far from professional forecasts (Coibion, Gorodnichenko and Kumar, 2018). Simple communication reaches the public better than technical statements (Haldane and McMahon, 2018).

4. Inflation at risk

Policymakers care about the risk of very high or very low inflation, not only the mean. Inflation-at-risk estimates the conditional quantiles of future inflation, \(Q_\tau(\pi_{t+h}\mid X_t)\), by quantile regression, analogous to growth-at-risk (Adrian, Boyarchenko and Giannone, 2019; López-Salido and Loria, 2024). Inflation persistence is typically stronger in the upper tail: when inflation is high, the risk of further increases rises disproportionately. Inflation targeting and central-bank independence compress the predictive distribution, but their additional predictive power is limited once current inflation is known, and they offered limited protection against the common 2021–23 shock. Exercise 15.2 estimates a pooled cross-country version.

5. Operating frameworks and the balance sheet

Central banks implement the policy rate either in a corridor system with scarce reserves, steering the overnight rate between a lending and a deposit facility through open-market operations, or in a floor system with abundant reserves, where the rate paid on reserves sets the floor (the Fed since 2008, ECB since 2015). Large balance sheets after QE have fiscal implications: when policy rates rise, interest paid on reserves exceeds income on long-term assets, and central banks such as the Fed and the Bank of England have reported operating losses since 2022. Quantitative tightening reduces the balance sheet gradually.

6. Lessons from the 2021–23 inflation

The inflation surge followed pandemic demand, supply-chain disruptions, energy price shocks after Russia's invasion of Ukraine, and large fiscal transfers. Central banks initially judged it transitory, then raised rates at the fastest pace in four decades. Lessons that recur in reviews: inflation can rise quickly and nonlinearly when labour markets are tight and shocks are large (Benigno and Eggertsson, 2023); forward guidance that commits to low rates can delay the response; make-up strategies designed for the lower bound were ill-suited to an inflation surge (the Fed removed its average-inflation-targeting language in 2025); and anchored expectations and credibility made disinflation less costly. Bernanke and Blanchard (2023) decompose U.S. inflation into contributions from energy and food prices, shortages and labour-market tightness.

7. Monetary policy in emerging and developing economies

Emerging-market central banks face fiscal dominance, dollarisation, shallow financial markets and volatile capital flows. Many combine interest-rate policy with foreign-exchange intervention and macroprudential tools (Lecture 8). Bangladesh's shift from monetary targeting to an interest-rate corridor (2023), a crawling peg (2024) and a market-based exchange rate (2025) illustrates the transition towards modern frameworks under high inflation and reserve pressures. Korea's experience since 1998 shows how inflation targeting with a flexible exchange rate can lower inflation and pass-through.

8. Course synthesis

The course moved from the quantity theory to the New Keynesian model, optimal policy, open economies, financial frictions, the lower bound, fiscal interactions, heterogeneity, estimation and identification. A good policy analysis combines a model for mechanisms and counterfactuals (Lectures 3–12), estimation that disciplines the model with data (Lecture 13), credible empirical evidence on transmission (Lecture 14), and institutional knowledge of how policy is made (this lecture).

W = readtable('cpi_inflation_wdi.csv');
codes = unique(W.code);
% TODO: for each country: adoption year (or 1998 for non-IT), pre and post averages; regression with ols_nw(..., 0)

W = readtable('cpi_inflation_wdi.csv');
codes = unique(W.code); R = [];
for c = 1:numel(codes)
  k = strcmp(W.code, codes{c}); yr = W.year(k); p = W.cpi_inflation(k); ity = W.it_year(find(k, 1));
  isit = ~isnan(ity) && ity >= 1990 && ity <= 2016;
  if ~isnan(ity) && ~isit, continue; end            % late or very recent adopters: drop
  t0 = 1998; if isit, t0 = ity; end
  pre = p(yr >= t0 - 5 & yr <= t0 - 1); post = p(yr >= t0 + 1 & yr <= t0 + 5);
  if numel(pre) < 5 || numel(post) < 5 || any(isnan([pre; post])) || mean(pre) > 40, continue; end
  R = [R; isit mean(pre) mean(post)];
end
d = R(:, 3) - R(:, 2);
for g = [1 0]
  m = R(:, 1) == g;
  if g, nm = 'IT adopters  '; else, nm = 'non-adopters '; end
  fprintf('%s n = %3d   before %5.1f%%   after %5.1f%%   change %+5.1f\n', nm, sum(m), mean(R(m, 2)), mean(R(m, 3)), mean(d(m)));
end
[b1, s1] = ols_nw(d, [ones(size(d)) R(:, 1)], 0);
[b2, s2] = ols_nw(d, [ones(size(d)) R(:, 1) R(:, 2)], 0);
fprintf('\nchange = %.2f %+.2f IT                       (IT s.e. %.2f)\n', b1(1), b1(2), s1(2));
fprintf('change = %.2f %+.2f IT %+.2f pre-inflation    (IT s.e. %.2f)\n', b2(1), b2(2), b2(3), s2(2));

IT adopters   n =  38   before  10.2%   after   4.7%   change  -5.5
non-adopters  n =  94   before   9.2%   after   4.6%   change  -4.6

change = -4.64 -0.85 IT                       (IT s.e. 1.32)
change = 1.25 -0.24 IT -0.64 pre-inflation    (IT s.e. 0.78)

W = readtable('cpi_inflation_wdi.csv');
% TODO: build pi_t and pi_{t+1} pairs within countries; IT = year >= it_year; qreg at tau = 0.1, 0.5, 0.9

W = readtable('cpi_inflation_wdi.csv');
codes = unique(W.code); Y = []; X = [];
for c = 1:numel(codes)
  k = find(strcmp(W.code, codes{c})); yr = W.year(k); p = W.cpi_inflation(k); ity = W.it_year(k(1));
  for t = 1:numel(yr) - 1
    if yr(t) < 1985 || yr(t+1) ~= yr(t) + 1 || isnan(p(t)) || isnan(p(t+1)) || abs(p(t)) > 50 || abs(p(t+1)) > 50, continue; end
    it = ~isnan(ity) && yr(t) >= ity;
    Y = [Y; p(t+1)]; X = [X; 1 p(t) it it * p(t)];
  end
end
fprintf('observations: %d (country-years), IT share %.0f%%\n\n', numel(Y), 100 * mean(X(:, 3)));
taus = [0.1 0.5 0.9]; Bq = zeros(4, 3);
fprintf('quantile    const    pi_t     IT     IT*pi_t\n');
for j = 1:3
  Bq(:, j) = qreg(Y, X, taus(j));
  fprintf('  %.1f     %6.2f   %6.3f  %6.2f   %6.3f\n', taus(j), Bq(:, j));
end
fprintf('\npredicted quantiles of next year''s inflation (%%)\n');
fprintf('current infl   framework    q10     q50     q90\n');
for p0 = [3 8]
  for it = [0 1]
    q = [1 p0 it it * p0] * Bq;
    if it, nm = 'IT     '; else, nm = 'non-IT '; end
    fprintf('     %d%%        %s   %6.2f  %6.2f  %6.2f\n', p0, nm, q);
  end
end

observations: 6098 (country-years), IT share 13%

quantile    const    pi_t     IT     IT*pi_t
  0.1      -0.40    0.309    0.35    0.081
  0.5       1.02    0.709    0.04   -0.060
  0.9       4.36    1.060   -1.67   -0.055

predicted quantiles of next year's inflation (%)
current infl   framework    q10     q50     q90
     3%        non-IT      0.53    3.14    7.54
     3%        IT          1.13    3.00    5.70
     8%        non-IT      2.08    6.69   12.83
     8%        IT          3.08    6.24   10.72