Forecast LabBangladesh economy · in numbers
LIVE FORECASTSUPDATED NOV 2026|4 SERIES · 80% RANGES · SCORED ON FIRST OFFICIAL RELEASES|DATA PUBLISHED 17 NOV 2026

What's next for Bangladesh's economy

This page publishes forecasts for four key measures of Bangladesh's economy: consumer-price inflation, GDP growth, workers' remittances and electricity demand. We compare a wide range of forecasting methods (statistical, econometric, machine-learning and combinations of them) and publish the one that holds up best, with an honest range around it.

Forecasts last updated . Official sources are checked every day.

We show how uncertain each forecast is and keep score against what actually happens. Forecasts are selected using out-of-sample validation and monitored for bias and stability.

Live track record
Every forecast is scored against the first official release. Results appear under "How we did" as new data arrive.
Actual Our forecast 80% range Ringed point = headline forecast · hover or tap a chart for the numbers
Consumer prices · yearly change
Inflation
About 8.3%
Likely to stay near this level to Apr 2027
Frequency
Monthly
Looking ahead
6 months, to Apr 2027
Latest actual
8.27% · Oct 2026
Source
BBS / BB
Likely range (80%)Nov 20267.3–9.2%Dec 20267.3–9.8%Jan 20276.8–10.2%Feb 20276.6–10.7%Mar 20276.7–10.9%Apr 20276.5–11.2%
Published 12 Nov 2026Headline: Jan 2027 · 3rd month ahead
Prices look set to keep rising at around 8% a year over the next several months. We don't see quick relief.
We tested 34 forecasting methods on monthly data back to 2013, together with models built on the taka exchange rate, world food and energy prices and domestic market prices of staple foods. None forecast better than assuming inflation stays at its latest reading, so that is our forecast. The months ahead show the range inflation is likely to fall in, which widens the further ahead we look.
Real GDP · yearly change
Economic growth
3.7%
Nowcast for FY27 Q1 (Jul–Sep 2026)
Frequency
Quarterly
Quarter in focus
Jul–Sep 2026
Latest actual
4.6% · FY26 Q4
Source
BBS / BB
By quarterFY27 Q1 (Jul–Sep 2026)3.7%
Published 16 Nov 2026Headline: FY27 Q1 (Jul–Sep 2026)
Growth looks set to stay moderate this quarter. This is a nowcast, an estimate made before the official figure, and we'll revisit it when new data arrive.
Actual growth figures are the Bangladesh Bureau of Statistics first quarterly estimates; FY26 Q4 (Apr–Jun 2026) was released on 7 October 2026.
Workers' remittances
Money sent home
$2.97bn
Forecast for Jan 2027
Frequency
Monthly
Looking ahead
3 months, to Jan 2027
Latest actual
$2.87bn · Oct 2026
Source
Bangladesh Bank
Month by monthNov 2026$2.89bnDec 2026$2.91bnJan 2027$2.97bn
Published 12 Nov 2026Headline: Jan 2027 · 3rd month ahead
Migrant workers are likely to keep sending home about US$2.9 billion a month through Jan 2027, with the usual swings around festivals and one-off events.
Power demand
Electricity demand
8.12 TWh
Forecast for Nov 2026
Frequency
Monthly
Looking ahead
6 months, to Feb 2027
Latest actual
10.77 TWh · Aug 2026
Source
Ember / BPDB
Month by monthSep 202610.47 TWhOct 20269.87 TWhNov 20268.12 TWhDec 20267.42 TWhJan 20277.92 TWhFeb 20277.74 TWh
Published 20 Oct 2026Headline: Nov 2026 · 3rd month ahead
Demand should stay near 10 TWh through October, then ease into the winter months, as it does every year.

Our forecasts, on the record

One current forecast for each series. We score every one against the first official release, and we never edit a forecast after publishing it.

1 current forecast per series
SeriesPublishedForecast80% rangeActual (first release)ErrorStatus
Inflation · Jan 202712 Nov 20268.27%6.80–10.18%——Waiting for the official number
Economic growth · FY27 Q1 (Jul–Sep 2026)16 Nov 20263.7%1.52–5.59%——Waiting for the official number
Money sent home · Jan 202712 Nov 2026$2.97bn$2.55–3.48bn——Waiting for the official number
Electricity demand · Nov 202620 Oct 20268.12 TWh7.64–8.97 TWh——Waiting for the official number
ForecastThe number we published before the official figure was out.
ActualThe first official release for the same period. Later revisions are kept separately.
ErrorForecast minus actual. The smaller, the better.

What changed

A dated note each time the forecasts are updated, including what moved and by how much.

5 updates logged
  1. New official data: Economic growth for FY26 Q4 (Apr–Jun 2026)
    • Inflation: no change.
    • Economic growth, FY27 Q1 (Jul–Sep 2026): 3.7% (was 4.2%).
    • Money sent home: no change.
    • Electricity demand: no change.
  2. New official data: Inflation for Oct 2026; Money sent home for Oct 2026
    • Inflation, Jan 2027: 8.27% (was 8.34%).
    • Economic growth: no change.
    • Money sent home, Jan 2027: $2.97bn.
    • Electricity demand: no change.
  3. New official data: Electricity demand for Aug 2026
    • Inflation: no change.
    • Economic growth: no change.
    • Money sent home: no change.
    • Electricity demand, Nov 2026: 8.12 TWh (was 8.08 TWh).
  4. New official data: Inflation for Sep 2026
    • Inflation, Dec 2026: 8.34% (was 8.26%).
    • Economic growth: no change.
    • Money sent home: no change.
    • Electricity demand: no change.
  5. Forecasts rebuilt after a full comparison of methods
    • Economic growth, FY27 Q1 (Jul–Sep 2026): 4.2% (was 4.9%). Growth history for FY24 and FY25 was also aligned with the official series.
    • Money sent home, Dec 2026: $2.92bn (was $3.44bn). Forecasts now stop at three months, as far as testing supports.
    • Inflation: central forecast unchanged at 8.26%. Ranges widened about twofold after testing showed the earlier ones were too narrow.
    • Electricity demand: no change.
Need a forecast built for your organization? Mangrove Intelligence builds commissioned forecasts, scenario analysis and forecasting systems for economic, financial, business and operational data.
Talk to Mangrove Intelligence
How the forecasts are madeMethods, data and scoring

What we publish

One central estimate for each series, with an 80% range around it. The range is where we expect the actual figure to land about four times out of five. It will sometimes land outside, and extreme events can fall well beyond it.

How we choose a forecast

We test many approaches on data they had not seen, stepping forward through history the way a real forecaster would. We publish the approach that performs best and stays stable. If nothing clearly beats a simple benchmark, we publish the simple benchmark. Being complicated is never a reason to choose a method.

Watching for bias

We track whether forecasts run systematically high or low, and we do not quietly adjust them to offset past misses. A method that leans one way does not make the cut just because its average error looks smaller.

Honest ranges

Ranges are built from our own past errors, and widened where history shows they would have been too narrow. We only look as far ahead as our track record supports, which is why remittances stop at three months.

The methods we compare

For every series we run the same head-to-head test across a wide library of methods, grouped here by family.

  • Simple benchmarks. The latest value carried forward, the same period a year earlier, the average recent trend, and moving averages over recent months or quarters. Every other method has to beat these.
  • Exponential smoothing. Simple, damped-trend and seasonal versions that give more weight to recent observations.
  • Theta method. A robust time-series method that blends the long-run trend with short-run movements.
  • ARIMA family. Autoregressions and ARIMA models, including seasonal versions.
  • Trend and calendar regressions. A trend plus month-of-year effects, with a short-term correction for recent errors. For remittances, also the timing of the Eid festivals, which moves from year to year.
  • Seasonal growth projections. Last year's value grown at the recent year-on-year rate.
  • Regularised regression. Ridge and elastic-net regressions on recent history, which guard against over-fitting.
  • Machine learning. Random forests and gradient boosting on recent history and calendar effects.
  • Models with outside indicators. The same machine-learning methods with World Bank commodity-price indicators (oil, energy and food) for inflation, remittances and electricity, and mixed-frequency nowcasts for GDP that use early remittance and electricity signals.
  • Forecast combinations. The median, a trimmed average, an average of the recent best performers, and an error-weighted blend of all of the above.

We add methods as the data allow. Not yet tested: multivariate systems (VAR and Bayesian VAR), dynamic factor models, mixed-frequency regressions (MIDAS), other gradient-boosting libraries such as XGBoost and LightGBM, and neural networks. They are on the list for when longer histories are available.

What each forecast measures

Inflation

Measures
Headline consumer-price inflation: the year-on-year change in the national consumer price index (all items), in %.
Source
Bangladesh Bureau of Statistics, published through Bangladesh Bank.
History
Monthly from July 2013. Earlier price-index bases are linked to the current 2021-22 base so the series is continuous.
Looks ahead
Up to six months.
In use today
The latest reading carried forward. No tested method beat it reliably, so we publish the simple benchmark, with a wider range because past ranges were too narrow.
Inputs tested
The series' own history, seasonality and World Bank commodity prices.

Economic growth

Measures
Real GDP growth: the year-on-year change in constant-price GDP, in %, by fiscal quarter (Q1 is July to September).
Source
BBS national accounts, published through Bangladesh Bank.
History
Quarterly from FY2016-17 Q1. The latest quarter, FY26 Q4, is the BBS first estimate released on 7 October 2026.
Looks ahead
One quarter. This is a nowcast: an estimate made before the official figure is out.
In use today
The average of the latest four quarters of growth. Models using early remittance and electricity signals did not beat it.
Inputs tested
The series' own history, and remittance and electricity growth as early signals.

Money sent home

Measures
Workers' remittance inflows, in US$ million per month.
Source
Bangladesh Bank (Monthly Economic Trends). The latest month can be preliminary until Bangladesh Bank publishes it.
History
Monthly from July 2014.
Looks ahead
Up to three months, as far as our track record supports.
In use today
A weighted blend of several statistical methods, which edged out the best simple benchmark.
Inputs tested
The series' own history, seasonality, Eid timing, and World Bank oil and energy prices.

Electricity demand

Measures
Monthly electricity demand, in terawatt-hours (TWh).
Source
Ember monthly electricity data, cross-checked against BPDB generation data where available.
History
Monthly from January 2019.
Looks ahead
Up to six months.
In use today
A trend-and-seasonal regression with a short-term correction. It beat the best simple benchmark by a wide margin at every horizon.
Inputs tested
The series' own history, seasonality and World Bank energy prices.

Reading the dates

Monthly figures (inflation, remittances, electricity) are labelled by the calendar month they describe. Growth is quarterly and follows Bangladesh's fiscal year, which starts in July: Q1 is July to September, Q2 October to December, Q3 January to March and Q4 April to June. So FY27 Q1 means July to September 2026. "Looking ahead" counts from the latest official figure, so a month that has ended but has not been released yet is still a forecast. Each card lists the full path, month by month; the ringed point and the large number are the headline forecast. For growth, the quarter that has just ended is a nowcast: an estimate made before the official figure is out.

How a forecast is tested

  • Out of sample, step by step. We go back through history in expanding windows (rolling origin). At each step a method sees only the data that existed at that point, forecasts the following months or quarters, and is scored against what then happened. Calendar effects such as Eid are known in advance, so they are allowed. Nothing from the future leaks in.
  • Error and loss. Error is forecast minus actual, so a positive error means we were too high. Methods are ranked on mean absolute error, which is not dominated by a few extreme misses. Squared-error measures are reported but do not decide the ranking.
  • Enough tests. A method needs at least 24 test points for a monthly series and 14 for GDP, and cannot fail to produce a forecast more than 10% of the time. In practice we have around two dozen tests for GDP, 50 to 90 for electricity and remittances, and about 100 for inflation.
  • Each horizon separately. One month ahead and six months ahead are different problems, so every horizon is tested and chosen on its own.

How the published method is chosen

The selection rule was written down and frozen before we looked at the results, so the winner is not fitted to the scoreboard.

  • Bias screen. A method is dropped if its average error is large relative to its typical error (above 0.40 of it), or if it misses in the same direction more than 70% of the time.
  • Simple unless proven otherwise. If the best method is within 3% of a simpler one, we take the simpler one. A more complex method must beat the best simple benchmark in a one-sided Diebold–Mariano test (Harvey–Leybourne–Newbold small-sample version) at the 10% level, and must not be worse than the benchmark by more than 5% in either half of the test period.
  • Combinations next. If no single method is clearly better, we publish the best combination of methods when it is at least as accurate as the benchmark and stable across both halves.
  • Otherwise the benchmark. If none of that holds, the best simple benchmark is the forecast.
  • Disagreement is information. If the best methods disagree widely about today's forecast (a spread of more than twice the selected method's typical error), we flag the forecast as fragile instead of averaging the disagreement away.

After seeing the first results we made three refinements, which we record here because they came after the fact: prefer the same method across horizons when accuracy differs by under 2%; keep one method across all horizons when that gives a coherent path (used for inflation); and calibrate ranges across horizons together rather than one at a time. We will re-validate these choices once each series has about six scored forecasts.

How the ranges are built

The 80% range is the forecast plus the 10th and 90th percentiles of that method's own earlier errors at the same horizon, with at least 12 errors needed. We do not assume errors follow a bell curve, so a range can be lopsided. Where testing showed such ranges were too narrow, we widen them by a common factor, capped at 2. Inflation ranges are widened to twice their raw width, which brought test coverage to roughly 72% to 83% across horizons. A horizon is published only if coverage in testing falls between 70% and 90% and the method is no worse than the benchmark. That is why remittances stop at three months and growth at one quarter.

The limits of forecasting

We take seriously the critique of economic forecasting that single-number predictions overstate what is knowable, especially for variables with rare, very large moves. We do not treat any one view as the final authority on which method to use. We turn the useful lessons into rules.

  • A point forecast is not knowledge. We always publish a range with the number, and we avoid false precision.
  • Errors are not bell-shaped. Growth and remittance changes have fat tails: rare, very large misses happen. Ranges come from actual past errors, not from a normal-distribution formula.
  • A range is not a worst case. The 80% range will miss about one time in five, and a real shock can land far outside it. We track the largest past miss separately as a stress reference and never present it as a probability.
  • Robustness over a fragile gain. A method that is slightly more accurate on average but leans one way, or only works in half the history, does not win.
  • No automatic bias adjustment. We report bias but never quietly shift a forecast to offset past misses. An adjustment would have to pass its own out-of-sample test, use only earlier errors, be shrunk towards zero and be capped.
  • Shorten the horizon when evidence weakens. We would rather say less than claim what the record cannot support.
  • Archive every miss. Every forecast stays on the record and is scored.

What the tests showed

  • Electricity: trend-and-seasonal models beat the best simple benchmark by roughly a third at every horizon, a clear and statistically solid gain.
  • Inflation: no method reliably beat carrying the latest reading forward, so that is what we publish, with ranges widened because the raw ones were too narrow.
  • Remittances: no single method significantly beat a three-month average. A blend of methods was slightly better. Projecting last year's seasonal pattern forward was clearly worse.
  • Growth: nothing beat the average of the latest four quarters, including models using early remittance and electricity signals. Growth has had occasional very large misses, so we treat it as a nowcast and keep the range wide.
  • Outside indicators: World Bank oil, energy and food prices added no accuracy out of sample. Longer yearly World Bank histories (decades of inflation, growth and remittances) pointed the same way: simple methods were hard to beat.

Honest limits: histories are short (electricity starts in 2019, quarterly growth in 2016-17), tests use the latest revised data rather than what was first published, and the live record is only beginning. Scoring against first official releases is the real test, and it is what this page will show over time.

Data and scoring

We use official releases: the Bangladesh Bureau of Statistics and Bangladesh Bank for prices, growth and remittances, and Ember for electricity. Older price-index bases are linked to the current base so inflation forms one continuous series. When BBS has published an annual growth figure but not yet the matching quarter, that quarter is implied from the annual figure and marked with an asterisk until the quarterly release arrives. The latest month of remittances can be preliminary. World Bank commodity-price and annual data are used for testing outside indicators. We score each forecast against the first official number. Later revisions are kept separately and never rewrite the score.

Updates and the record

Sources are checked every day, and the forecasts change only when new official data arrive. Automatic checks flag gaps, implausible jumps and large revisions, and every update is reviewed by hand before it goes live. Each version is archived with its date and never edited, and every miss stays on the record.

Reading the charts. The light line is what has happened. The coloured line is our central estimate. The shaded band is the range we expect the outcome to land in about four times out of five.