## wpiea2022183-print-pdf - REFERENCES

## Source details

**Canonical URL:** [wpiea2022183-print-pdf - REFERENCES](https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022183-print-pdf.pdf)

## Other formats

- [Markdown version](/-/media/files/publications/wp/2022/english/wpiea2022183-print-pdf.pdf.md)
- [Structured JSON version](/-/media/files/publications/wp/2022/english/wpiea2022183-print-pdf.pdf.json)

---

### I. Introduction — research scope and objectives
- Research question: What has been the role of the Organization of the Petroleum Exporting Countries (OPEC) in the oil market and how has that role evolved over time (analysis period: 1988–2019 for price effects)?
- Objectives:
  - Test OPEC’s ability to influence oil price levels and volatility by examining effects of OPEC meetings on oil prices.
  - Study determinants of OPEC production decisions.
  - Discuss recent developments including the OPEC+ alliance.
- Key contextual points:
  - OPEC’s stated objective: coordinate petroleum policies to stabilize oil markets around a “fair” price.
  - OPEC’s organization is fragile and lacks a formal enforcement mechanism for quotas.
  - OPEC accounted for more than 40 percent of world oil production over the last three decades.

### II. Event-study methodology and data definitions
- Sample and meetings:
  - Analysis covers 101 meetings from 1987 to 2019 (From June 1987 to July 2019).
  - Regular meetings: 71; non-regular meetings: 30.
  - Regular meetings usually last two days and conference resolutions become effective after 30 days.
  - Non-regular meetings are usually announced 3-5 days before their start.
- Decision categories and counts:
  - Neutral decisions: 58 total (42 Regular and 16 Non-Regular).
  - Cut decisions: 24 total (14 Regular and 10 Non-Regular).
  - Boost decisions: 19 total (15 Regular and 4 Non-Regular).
- Event-window definitions and return construction:
  - Cumulative daily oil return R_k,j defined over a window T where T = 11 (an 11 trading day asymmetric window is used).
  - Cumulative returns prior to meeting defined with the opposite sign; j = 1, ..., T.
- Robustness checks:
  - Also used excess returns (residuals from regressing daily oil price log-change on S&P500 total returns) and Brent 3-month futures daily prices (available since 1990).

### III. Findings on predictability and market reaction
- Predictability and surprises:
  - OPEC’s decisions are not exogenous; markets can anticipate decisions to some extent.
  - The surprise (unexpected) component of OPEC decisions is the element that moves prices.
  - No systematic bias: oil returns after events are not systematically related to the decision type (cut, maintain, boost).
- Multinomial logit model performance:
  - Model can predict the right outcome 2/3 of the times (66 percent correct; sample N = 95 meetings in econometric sample).
  - McFadden pseudo R2 = 0.16.
  - Empirical classification (sample N = 95):
    - cut: 23 occurrences, 24.2 percent, model correct 39 percent
    - neutral: 56 occurrences, 59.0 percent, model correct 93 percent
    - boost: 16 occurrences, 16.8 percent, model correct 13 percent
- Main drivers identified:
  - Cyclical component of oil prices is the most significant predictor of meeting outcomes; trend component is insignificant.
  - Economic uncertainty increases the probability of a cut.
  - Entering a meeting with a low Saudi oil market share reduces the probability of an (extra) cut.
- Volatility and abnormal returns:
  - Oil market volatility is abnormally high around OPEC meeting dates, including prior to meetings, consistent with leaks and rumors.
  - The day-after announcement volatility is higher than the median volatility in the control sample by:
    - 2.2 percentage points higher for regular meetings.
    - 3.1 percentage points higher for non-regular meetings.
  - On average, OPEC tends to be a stabilizing force: market volatility drops below its median value in the control sample about 9-10 days after meetings, especially for non-regular meetings.
- Market stabilization nuance:
  - Not all meetings stabilize prices; episodes exist where OPEC aimed to regain market share rather than stabilize prices (e.g., second half of the 1990s).
  - Leaks and exceptional episodes (counter-oil shock 1986, breakdown in November 2014, 2020 price war) can strongly move the market.

### IV. Formal representation of impacts (notation and interpretation)
- Key modelization:
  - Return representation: R_k,t = g_t(x_k,t) + γ_t y_k,t where x are market fundamentals, y are OPEC decisions, γ_t captures price reaction to OPEC’s decision.
  - Forecast error expression highlights that only the unexpected component of y_k moves markets.
  - If γ_t > 0, volatility increases around meeting dates unless decisions are perfectly predictable.
- Statistical windowing logic:
  - Narrow windows (T = 11 trading days) reduce the contribution of changing fundamentals to observed abnormal returns.

### V. Market responses to OPEC announcements — event-window evidence
- Event window choices:
  - Used 3 days window; 11 days appears appropriate to identify market behavior before OPEC meetings.
  - 3 days window reflects how the market behaves for OPEC’s announcement of the irregular meeting.
- Volatility comparisons and distributional context:
  - Difference measured as standard deviation of the log Brent price post minus pre meeting using asymmetric 11-days window; 3-lags moving average shown.
  - The 10th and 90th percentile of the Brent oil return distribution is 1.8 and 3.1 percent, respectively.
  - Excluding extreme events in 2020, oil price return distribution is slightly skewed to the right.
- Regular meetings:
  - Volatility of oil price daily returns increases as the conclusion of regular meetings approaches.
  - Day three and day one before the meeting show higher oil return volatility than the typical volatility of the control distribution (about 0.6 percentage points higher than the median volatility).
  - The day before the start of the meeting (day -2) shows unusually low volatility (about 0.5 percentage point below the median volatility).
  - The day after the concluding meeting volatility peaks, implying OPEC decisions move the market and are not always fully anticipated.
  - After the first market reaction, volatility declines afterward.
- Non-regular meetings:
  - Volatility of oil return in the days before non-regular meetings is almost twice as high as the typical median volatility of oil returns.
  - After the meeting, the price volatility is abnormal about 3.5 percent—i.e., 1.3 percentage points above the median volatility.
  - As days pass, volatility falls substantially—from above the 75th percentile to below the 25th percentile of the control distribution.
- Overall summary:
  - Prices fluctuate more than typical even before meeting conclusions (likely due to leaks and rumors), the day after concluding meetings shows higher volatility, and over time volatility is reduced, especially for non-regular meetings, suggesting stabilizing effects after initial impact.

### VI. Effectiveness of decisions on prices — return patterns
- Daily return behavior and cumulative returns:
  - Average daily return is typically quite noisy around meeting dates.
  - After a production cut the oil return increases on average; after a production boost it declines, relative to previous day.
- Regular meetings (dynamics):
  - Production boosts are typically preceded by an upward price trajectory that the boost does not meaningfully alter; price appears to stabilize only toward the end of the time window.
  - Production cuts are preceded by falling prices that the cut typically does not halt.
  - In both cases there is an initial effect the day after the meeting but no long-lasting impact on prices.
  - Unchanged production is typically related to slightly falling prices prior to meetings and reduced oil returns after meetings (examples: November 2014, March 2020).
- Non-regular meetings (dynamics):
  - Announcements occur a few days before the meeting, so some price effect happens before the conclusion.
  - For both boosts and cuts, price impact is visible at least 6 days before the concluding meeting.
  - The price impact is substantial (about 3 percent) but not long-lasting, suggesting market overreaction to meeting announcements that fades as the meeting approaches.
- Interpretation:
  - Some events with statistically insignificant price response are attributed to well anticipated decisions.
  - Some decisions may have been perceived as not credible; lack of credibility likely relates to low degree of compliance with production quotas.

### VII. Compliance, credibility, and behavior
- Tradeoffs and heterogeneity:
  - OPEC decisions involve a tradeoff between supporting/stabilizing price and maintaining market share; members differ by dependence on oil, spare capacity, fiscal position, business and political cycle stage, inflation, exchange rate regime, and reserves.
  - Small producers face temptation to free ride due to no explicit enforcement mechanism.
  - Saudi Arabia has usually acted as swing producer—offsetting over- and under-compliance of other members—bearing asymmetric benefits and costs.
  - Saudi Arabia was over compliant during the study period and demonstrated higher discipline than other OPEC members.
- Formal measure of coalition compliance:
  - Overall OPEC compliance level, φ, defined as
    - φ = 100 * ( (∑ Production_i) / (∑ Allocation_i) − 1 ), where n is the number of members in the coalition.
    - If φ > 0 (φ < 0) there is under (over) compliance at the coalition level.
- Historical patterns and observations:
  - 1980s: compliance deteriorated due to geopolitical tensions.
  - 1994–1999: compliance reverted to stability.
  - 2011–2014: compliance declined given strong growth in oil demand; subsequently contributed to a supply glut as US shale growth surprised on the upside.
  - No strong relation between periods of higher compliance and market volatility in the sample.
- Data and scope notes:
  - Figures and tables show average compliance behavior by decade; allocation and production are based on historical composition and OPEC membership assessed in April 2021 (note excludes Libya in some contexts).
  - Table data represent averages per decade (1984-1989, 1990-1999, 2000-2009, 2010-2019) and reflect current members with noted country-specific adjustments (e.g., Ecuador membership history).

### VIII. Drivers of OPEC decisions: framework, variables, and magnitudes
- Theoretical framework:
  - Stylized equilibrium model (Nakov and Pescatori (2010)) with a dominant producer and a competitive fringe.
- Candidate factors considered:
  - Oil demand conditions, oil demand outlook, and forecast uncertainty.
  - OPEC market power: low OPEC share of global production signals reduced market power and lower probability of a cut.
  - Real-time indicators: oil prices and US or OECD oil stocks as proxies for current and expected market tightness.
- Econometric specification:
  - Ordered multinomial logit with y = 0 (cut), 1 (keep production as is), 2 (boost).
  - Regressors include 12-month ahead forecast of US GDP growth and forecast dispersion (dispersion orthogonalized relative to US GDP forecasts), OECD stocks, Saudi Arabia share of oil production, cyclical and trend components of log real Brent price (HP filter), controls: AAA-spread, US T-bill rate.
- Key magnitude results:
  - Saudi share effect:
    - One standard deviation decline in Saudi share (about 1 percentage point) decreases the probability of a cut by 0.10 and increases the probability of a production increase by 0.08.
  - Cyclical oil price effect:
    - One standard deviation increase in the cyclical component of oil prices (i.e., 17 percent) induces a 0.16 reduction in the probability of a cut and increases the probability of a boost by 13 percent.
  - Uncertainty effect:
    - One standard deviation increase in economic forecast uncertainty induces a 0.07 increase in the probability of a cut.
- Robustness:
  - Main conclusions remain under additional explanatory variables and shorter sample periods.
  - When oil prices are not decomposed, the time trend becomes significant.
  - Futures prices have less explanatory power than spot prices.
  - Saudi market share role is robust to starting the sample in the 2000s; economic uncertainty may lose significance with shorter samples.

### IX. OPEC communication — text analysis findings
- Corpus and methods:
  - Corpus: 58 OPEC meeting concluding statements (2002–2019), two Consultative Meeting statements, and 51 opening address statements.
  - Methods: cosine similarity and TF-IDF applied to a Bag of Words representation (Python).
- Preprocessing:
  - Remove stop words; stem words; apply TF-IDF weighting with formula w_{dj} = log( M / N_d + 1 ).
- Findings:
  - Transparency (informativeness) in statements has modest fluctuations; statements were less repetitive around the Global Financial Crisis, the 2008 price boom, and the 2010 price recovery.
  - Extraordinary meetings have less repetitive statements; average difference with regular meetings is not substantial.
  - Since OPEC+ establishment, OPEC concluding statements (released one day before OPEC+ statements) have become less informative, consistent with growing relevance of OPEC+ and Russia.
  - Statements rarely reference geopolitical or weather events and more often highlight supply conditions than demand conditions.

### X. OPEC+ — features, challenges, and implications
- Coalition features:
  - OPEC+ includes a group of non-OPEC exporters (including Russia) and represents about 60 percent of global crude oil production.
  - Governance and double leadership (KSA and Russia) add complexity and instability to decision-making.
- Notable episodes:
  - March 2020: A “price war” after Russia–KSA clash produced a >50 percent oil price collapse.
  - Institutional timing issue: OPEC concluding statements are released before OPEC+ meetings; OPEC statements can be contradicted by subsequent OPEC+ outcomes.
- Trade-offs and implications:
  - Higher market share increases the weight of decisions.
  - Coalition instability may reduce medium-term credibility.
  - Pandemic period: high compliance allowed OPEC+ to implement large cuts and stabilize the market; fuller analysis pending more data.

### XI. Event catalog, selected single-event values, and aggregate statistics
- Event counts and meeting types:
  - Total meetings: 101 meetings.
  - Non-regular (extraordinary) meetings: 30.
  - Regular OPEC meetings usually last two days.
- Selected single-event entries (exact values preserved):
  - 1987-06-29 — Boost — -0.013 — 0.017 — 0.029
  - 2003-07-31 — Neutral — -0.016 — 0.063 — 0.078
  - 2016-11-30 — (3-days Cumulative Return) 9.33% — (11-days Cumulative Return) 9.15%
  - 2000-09-11 — (3-days Cumulative Return) -14.50% — (11-days Cumulative Return) -13.48%
  - 1990-07-27 — (3-days Cumulative Return) 1.80% — (11-days Cumulative Return) 21.98%
- Aggregate summary statistics by decision type and decade (excerpted; exact numeric columns preserved in source tables):
  - Cut All: Number of Events 240 — Average 0.013 — Max 0.134 — Min -0.104 — Price Return Before Announcement 0.000 — After Announcement 0.095 — Difference -0.150 — Volatility -0.013
  - Boost All: Number of Events 19 — Average -0.019 — Max 0.060 — Min -0.131 — Price Return Before Announcement 0.013 — After Announcement 0.220 — Difference -0.135 — Volatility 0.032
  - Neutral All: Number of Events 580 — Average 0.003 — Max 0.158 — Min -0.091 — Price Return Before Announcement -0.004 — After Announcement 0.107 — Difference -0.161 — Volatility -0.007
  - All All: Number of Events 1010 — Average 0.001 — Max 0.158 — Min -0.131 — Price Return Before Announcement 0.000 — After Announcement 0.220 — Difference -0.161 — Volatility -0.001
- Event interpretation rule:
  - R denotes oil price cumulative return in the [3] and [11] days after the conclusion of the meeting.
  - If the decision was unexpected: a cut implies R>0, a boost R<0, and no change R=0.

### XII. Data appendix — sources and construction
- OPEC production and allocation:
  - Crude oil production data source: IEA MODS Platform.
  - Production data ranged from 1984 to 2020.
  - OPEC organization provides oil production data only from 2001.
  - U.S. Energy Information Administration (EIA) reports OPEC crude oil production data in quarterly, monthly, and annual forms, but only since 1994.
  - OPEC allocation data are taken from OPEC bulletin publications: 1999, 2005, and 2020.
- Oil price data:
  - Brent crude oil price data: 1985 to 2019 obtained from FRED.
  - "Three-month" Brent futures contracts data: 1988 to 2019 obtained from Bloomberg terminal DataStream.
  - Brent chosen as the European benchmark price to minimize regional bias relative to WTI and Dubai.
- Macroeconomic and composite variables:
  - World Economic Outlook (WEO) database used for macroeconomic data.
  - GDP forecast and GDP forecast standard deviation indices constructed from IMF consensus forecast database (1989 to 2019).
  - IEA MODS Platform database used from 1989 to 2019 to calculate OPEC market share and OECD total oil stock used as a measure of market tightness.
  - Macroeconomic controls from FRED: three-month U.S. Treasury bills, AAA Moody's Corporate Bond Yield, and a trade-weighted index included.

*Source: wpiea2022183-print-pdf (References and Data Appendix) — IMF staff material as provided in the content unit.*

### REFERENCES..............................................................................................................

### wpiea2022183-print-pdf - REFERENCES.............................................................................................................. 

### I. Introduction — Research scope and objectives
- Research question: What has been the role of the Organization of the Petroleum Exporting Countries (OPEC) in the oil market and how has that role evolved over time (analysis period: 1988–2019 for price effects)?
- Objectives:
  - Test OPEC’s ability to influence oil price levels and volatility by examining effects of OPEC meetings on oil prices.
  - Study determinants of OPEC production decisions.
  - Discuss recent developments including the OPEC+ alliance.
- Key contextual points:
  - OPEC’s stated objective: coordinate petroleum policies to stabilize oil markets around a “fair” price.
  - OPEC’s organization is fragile and lacks a formal enforcement mechanism for quotas.
  - OPEC accounted for more than 40 percent of world oil production over the last three decades.

### II. Event-study methodology and data definitions
- Sample and meetings:
  - Analysis covers 101 meetings from 1987 to 2019 (From June 1987 to July 2019).
  - Regular meetings: 71; non-regular meetings: 30.
  - Regular meetings usually last two days and conference resolutions become effective after 30 days.
  - Non-regular meetings are usually announced 3-5 days before their start.
- Decision categories and counts:
  - Neutral decisions: 58 total (42 Regular and 16 Non-Regular).
  - Cut decisions: 24 total (14 Regular and 10 Non-Regular).
  - Boost decisions: 19 total (15 Regular and 4 Non-Regular).
- Event-window definitions:
  - Cumulative daily oil return R_k,j defined over a window T where T = 11 (an 11 trading day asymmetric window is used).
  - Cumulative returns prior to meeting defined with the opposite sign; j = 1, ..., T.
- Robustness checks:
  - Also used excess returns (residuals from regressing daily oil price log-change on S&P500 total returns) and Brent 3-month futures daily prices (available since 1990).

### III. Findings on predictability and market reaction
- Predictability and surprises:
  - OPEC’s decisions are not exogenous; markets can anticipate decisions to some extent.
  - The surprise (unexpected) component of OPEC decisions is the element that moves prices.
  - Table 9 (as referenced) shows no systematic bias: oil returns after events are not systematically related to the decision type (cut, maintain, boost).
- Multinomial logit model:
  - Model estimates cut, neutral, and boost decisions.
  - The cyclical component of oil prices is the most significant predictor of meeting outcomes.
  - The trend component of the oil price is insignificant for decisions.
  - Economic uncertainty increases the probability of a cut.
  - Entering a meeting with a low Saudi oil market share reduces the probability of an (extra) cut.
  - The multinomial logit can predict the right outcome 2/3 of the times.
- Volatility and abnormal returns:
  - Oil market volatility is abnormally high around OPEC meeting dates, including prior to meetings, consistent with leaks and rumors.
  - The volatility of oil market returns before and after meetings is higher and fluctuates more than in the control sample.
  - The day-after announcement volatility is higher than the median volatility in the control sample by:
    - 2.2 percentage points higher for regular meetings.
    - 3.1 percentage points higher for non-regular meetings.
  - On average, OPEC tends to be a stabilizing force: market volatility drops below its median value in the control sample (and pre-meeting average) about 9-10 days after meetings, especially for non-regular meetings.
- Market stabilization nuance:
  - Not all meetings stabilize prices; there are episodes where OPEC aimed to regain market share rather than stabilize prices (e.g., second half of the 1990s).
  - Leaks and exceptional episodes (counter-oil shock 1986, breakdown in November 2014, 2020 price war) can strongly move the market.

### IV. Formal representation of impacts (notation and interpretation)
- Key modelization:
  - Return representation: R_k,t = g_t(x_k,t) + γ_t y_k,t where x are market fundamentals, y are OPEC decisions, γ_t captures price reaction to OPEC’s decision.
  - Forecast error expression highlights that only the unexpected component of y_k moves markets.
  - If γ_t > 0, volatility increases around meeting dates unless decisions are perfectly predictable.
- Statistical windowing logic:
  - Narrow windows (T = 11 trading days) reduce the contribution of changing fundamentals to observed abnormal returns.

### V. Compliance, behavior, and transparency findings
- Compliance:
  - OPEC member compliance to production agreements has fluctuated historically, affecting credibility in some periods.
- Text analysis:
  - Concluding-statement text analysis shows OPEC’s level of transparency has moderately fluctuated over time.
  - Fewer repetitive statements were found around the Global Financial Crisis, during the 2008 oil price boom, and during the 2010 oil price recovery.
  - Extraordinary meetings tend to have fewer repetitive statements than regular meetings, though the average difference is not significant.

### VI. Implications for OPEC+ and recent developments
- OPEC+:
  - The paper touches on OPEC+ and flags potential complexity when market dynamics normalize after COVID-19 (section VI provides discussion).
- Policy-relevant implications (drawn from findings):
  - Market participants partially anticipate OPEC decisions; transparency and systematic rules would alter the predictable/unpredictable balance.
  - The stabilizing role of OPEC is conditional: non-regular meetings often induce stronger price corrections than regular meetings.
  - Predictive model findings suggest OPEC reacts mainly to cyclical oil price components and economic uncertainty rather than trend shifts.

*Source: wpiea2022183-print-pdf - REFERENCES (IMF staff paper content provided in the supplied PDF).*

### 2.2 percent.

### 2.2 percent.

### Market responses to OPEC announcements
- Event window choices:
  - Used 3 days window; 11 days appears appropriate to identify market behavior before OPEC meetings.
  - 3 days window reflects how the market behaves for OPEC’s announcement of the irregular meeting.
- Volatility comparisons:
  - Difference measured as standard deviation of the log Brent price post minus pre meeting using asymmetric 11-days window; 3-lags moving average shown.
  - The 10th and 90th percentile of the Brent oil return distribution is 1.8 and 3.1 percent, respectively.
  - Excluding extreme events in 2020, oil price return distribution is slightly skewed to the right.
- Regular meetings:
  - Volatility of oil price daily returns increases as the conclusion of regular meetings approaches.
  - Day three and day one before the meeting show higher oil return volatility than the typical volatility of the control distribution (about 0.6 percentage points higher than the median volatility).
  - The day before the start of the meeting (day -2) shows unusually low volatility (about 0.5 percentage point below the median volatility).
  - The day after the concluding meeting volatility peaks, implying OPEC decisions move the market and are not always fully anticipated.
  - After the first market reaction, volatility declines afterward.
- Non-regular meetings:
  - Volatility of oil return in the days before non-regular meetings is almost twice as high as the typical median volatility of oil returns.
  - After the meeting, the price volatility is abnormal about 3.5 percent—i.e., 1.3 percentage points above the median volatility.
  - As days pass, volatility falls substantially—from above the 75th percentile to below the 25th percentile of the control distribution.
- Overall:
  - On average, OPEC has affected the oil market: prices fluctuate more than typical even before meeting conclusions (likely due to leaks and rumors), the day after concluding meetings shows higher volatility, and over time volatility is reduced, especially for non-regular meetings, suggesting stabilizing effects after initial impact.

### Effectiveness of decisions on prices
- Daily return behavior and cumulative returns:
  - Average daily return is typically quite noisy around meeting dates.
  - After a production cut the oil return increases on average; after a production boost it declines, relative to previous day.
  - For regular meetings:
    - Production boosts are typically preceded by an upward price trajectory that the boost does not meaningfully alter; price appears to stabilize only toward the end of the time window.
    - Production cuts are preceded by falling prices that the cut typically does not halt.
    - In both cases there is an initial effect the day after the meeting but no long-lasting impact on prices.
    - Unchanged production is typically related to slightly falling prices prior to meetings and reduced oil returns after meetings (examples cited: November 2014, March 2020).
  - For non-regular meetings:
    - Announcements occur a few days before the meeting, so some price effect happens before the conclusion.
    - For both boosts and cuts, price impact is visible at least 6 days before the concluding meeting.
    - The price impact is substantial (about 3 percent) but not long-lasting, suggesting market overreaction to meeting announcements that fades as the meeting approaches.
- Interpretation:
  - Some events with statistically insignificant price response are attributed to well anticipated decisions.
  - Some decisions may have been perceived as not credible; lack of credibility likely relates to low degree of compliance with production quotas.

### Compliance and credibility
- Tradeoff and heterogeneity:
  - OPEC decisions involve a tradeoff between supporting/stabilizing price and maintaining market share; member assessments vary by dependence on oil, spare capacity, fiscal position, business and political cycle stage, inflation, exchange rate regime, and reserves.
  - Small producers face temptation to free ride due to no explicit enforcement mechanism.
  - Saudi Arabia has usually acted as swing producer—offsetting over- and under-compliance of other members—bearing asymmetric benefits and costs.
  - Saudi Arabia was over compliant during the study period and demonstrated higher discipline than other OPEC members.
- Formal measure of coalition compliance:
  - Overall OPEC compliance level, φ, defined as
    - φ = 100 * ( (∑ Production_i) / (∑ Allocation_i) − 1 ), where n is the number of members in the coalition.
    - If φ > 0 (φ < 0) there is under (over) compliance at the coalition level.
- Historical patterns:
  - OPEC historical compliance varies across time and is influenced by global economic stages.
  - 1980s: compliance deteriorated due to geopolitical tensions.
  - 1994–1999: compliance reverted to stability.
  - 2011–2014: compliance declined given strong growth in oil demand; subsequently contributed to a supply glut as US shale growth surprised on the upside.
  - No strong relation between periods of higher compliance and market volatility in the sample.
- Data notes and scope:
  - Figures and tables show average compliance behavior by decade; allocation and production are based on historical composition and OPEC membership assessed in April 2021 (note excludes Libya in some contexts).
  - Table data represent averages per decade (1984-1989, 1990-1999, 2000-2009, 2010-2019) and reflect current members with noted country-specific adjustments (e.g., Ecuador membership history).

*Source: IMF staff analysis in wpiea2022183-print-pdf.*

### 2020. Indonesia suspended its membership in January 2009, reactivated it again in January 2016, but decided to suspend i

### wpiea2022183-print-pdf - 2020. Indonesia suspended its membership in January 2009, reactivated it again in January 2016, but decided to suspend i

### Drivers of OPEC decisions: framework and candidate factors
- The analysis is based on a stylized equilibrium model (Nakov and Pescatori (2010)) with a dominant producer and a competitive fringe.
- Candidate explanatory factor sets:
  - Oil demand conditions, oil demand outlook, and forecast uncertainty (a bleak outlook and elevated uncertainty should tend to increase the likelihood of a production cut).
  - OPEC market power: a low OPEC share of global production signals reduced market power and lower probability of a cut.
  - Real-time indicators: oil prices and US or OECD oil stocks as proxies for current and expected market tightness.
- Variables known at meeting time. Sample: all OPEC meetings from November 1989 to December 2018 for a total of 95 meetings.

### Econometric model specification
- Model: ordered multinomial logit with y = 0 (cut), 1 (keep production as is), 2 (boost).
- Probability formulation given by equation (8) in the source: p_dt = Pr(y_dt = i) = Pr(k_{i-1} < xβ + u ≤ k_i) = 1/(1+exp(−k_i + xβ)) − 1/(1+exp(−k_{i-1} + xβ)).
- Regressors (x) include:
  - 12-month ahead forecast of US GDP growth and forecast dispersion (forecast dispersion orthogonalized relative to US GDP forecasts).
  - OECD stocks.
  - Saudi Arabia share of oil production.
  - Cyclical and trend components of log real Brent price (Hodrick–Prescott filter).
  - Controls: AAA-spread, US T-bill rate (robustness checks include futures prices, contango, trade-weighted USD).

### Baseline results and predictive performance
- Goodness of fit:
  - McFadden pseudo R2 = 0.16.
  - Model sends a correct signal 66 percent of the times (sample N = 95 meetings).
- Empirical classification (Table 2 in source):
  - Actual decision frequencies and percent correct:
    - cut: 23 occurrences, 24.2 percent, model correct 39 percent
    - neutral: 56 occurrences, 59.0 percent, model correct 93 percent
    - boost: 16 occurrences, 16.8 percent, model correct 13 percent
    - Total: 95 meetings, model correct 66.3 percent
- Key statistical findings:
  - Oil prices deliver a fundamental role as indicators of market conditions; the cyclical component (not the trend) is the informative part.
  - Forecast dispersion (proxy for oil demand uncertainty) enters with a negative sign and is statistically significant at 3 percent.
  - High oil stocks increase probability of a cut, but when cyclical oil price is included, stocks are no longer significant.
  - Saudi production share (lagged one month) is significant and improves fit.

### Magnitudes and economic impacts (from baseline)
- Saudi share effect:
  - One standard deviation decline in Saudi share (about 1 percentage point) decreases the probability of a cut by 0.10 and increases the probability of a production increase by 0.08.
- Cyclical oil price effect:
  - One standard deviation increase in the cyclical component of oil prices (i.e., 17 percent) induces a 0.16 reduction in the probability of a cut and increases the probability of a boost by 13 percent.
- Uncertainty effect:
  - One standard deviation increase in economic forecast uncertainty induces a 0.07 increase in the probability of a cut.

### Robustness checks
- Main conclusions remain unaltered under additional explanatory variables and shorter sample periods.
- When oil prices are not decomposed, the time trend becomes significant.
- Futures prices have less explanatory power than spot prices (analysis constrained to 3-month futures due to data availability).
- Contango enters with expected sign but is not significant.
- Saudi market share role is robust to starting the sample in the 2000s; economic uncertainty may lose significance with shorter samples.

### OPEC communication: text analysis findings
- Corpus: 58 OPEC meeting concluding statements (2002–2019), two Consultative Meeting statements, and 51 opening address statements.
- Methods: cosine similarity and TF-IDF to assess repetitiveness/informativeness.
- Findings:
  - Transparency (informativeness) in statements has modest fluctuations; statements were less repetitive around the global financial crisis, the 2008 price boom, and the 2010 price recovery.
  - Extraordinary meetings have less repetitive statements; average difference with regular meetings is not substantial.
  - Since OPEC+ establishment, OPEC concluding statements (released one day before OPEC+ statements) have become less informative, consistent with growing relevance of OPEC+ and Russia.
  - Statements rarely reference geopolitical or weather events and more often highlight supply conditions than demand conditions.

### OPEC+ (expanded coalition) — features and implications
- OPEC+ includes a group of non-OPEC exporters (including Russia) and represents about 60 percent of global crude oil production.
- Governance and double leadership (KSA and Russia) add complexity and instability to decision-making.
- March 2020 example: A “price war” after Russia–KSA clash produced a >50 percent oil price collapse.
- Institutional timing issue: OPEC concluding statements are released before OPEC+ meetings; OPEC statements can be contradicted by subsequent OPEC+ outcomes.
- Trade-offs for OPEC+:
  - Higher market share increases the weight of decisions.
  - Coalition instability may reduce medium-term credibility.
- Pandemic period: high compliance allowed OPEC+ to implement large cuts and stabilize the market; fuller analysis pending more data.

### Event study and conclusions
- Event study results:
  - OPEC’s decisions are not systematically missed by markets, but surprise decisions induce sharp oil price movements regardless of direction (sizeable corrections driven by surprise).
  - Volatility of oil market returns before and after meetings is higher than typical (higher than volatility from a random sample of dates).
- Synthesis of main determinants:
  - Cyclical component of oil prices is the most significant factor in OPEC decisions; trend component is insignificant, implying OPEC stabilizes prices around a medium-term equilibrium rather than reacting to fundamental shifts.
  - Economic uncertainty increases probability of a cut.
  - Low Saudi oil market share significantly reduces probability of an (extra) cut.

*Source: wpiea2022183-print-pdf*

### References

### wpiea2022183-print-pdf - References

### Bibliographic references (selected)
- Adelman, M. A. (1980). Clumsy cartel. [Overcapacity/supply problems]. Energy J.;(United States), 1(1).
- Alm oguera, P. A., Dougla s, C. C., & Herrera, A. M. (2011). Testing for the cartel in OPEC: n on-cooperativ e co llu sion or just non-cooperative? Oxford Review of Economic Policy, 27(1), 144-168.
- Ball, L., & Sheridan, N. (2007). 6. Does Inflation Targeting Matter? (pp. 249-282). University of Chicago Press.
- Barsky, R. B., & Kilian, L. (2004). Oil and the macroeconomy since the 1970s. Journal of Economic Perspectives, 18(4), 115-134.
- Bernanke, B. S., & Mishkin, F. S. (1997). Inflation targeting: a new framework for monetary policy?. Journal of Economic perspectives, 11(2), 97-116.
- Bernanke, B. S., & Kuttner, K. N. (2005). What explains the stock market's reaction to Federal Reserve policy?. The Journal of finance, 60(3), 1221-1257.
- Burke, P. J., & Csereklyei, Z. (2016). Understanding the energy-GDP elasticity: A sectoral approach. Energy Economics, 58, 199-210.
- Brunetti, C., Büyükşahin, B., Robe, M. A., & Soneson, K. R. (2013). OPEC "fair price" pronouncements and the market price of crude oil. The Energy Journal, 79-108.
- Collier, P., & Venables, A. J. (2012). Greening Africa? Technologies, endowments and the latecomer effect. Energy Economics, 34, S75-S84.
- Csereklyei, Z., Rubio-Varas, M. D. M., & Stern, D. I. (2016). Energy and economic growth: the stylized facts. The Energy Journal, 37(2).
- Dargay, J., Gately, D. (1999). “Income's effect on car and vehicle ownership, worldwide: 1960–2015. Transportation Research Part A: Policy and Practice, 33(2), 101–138.
- Dargay, J., Gately, D., & Sommer, M. (2007). Vehicle ownership and income growth, worldwide: 1960-2030. The Energy Journal, 28(4).
- Demirgüç-Kunt, A., & Detragiache, E. (1998). The determinants of banking crises in developing and developed countries. Staff Papers, 45(1), 81-109.
- Draper D. W. (1984). The behavior of event-related returns on oil futures contracts. The Journal of Futures Markets (pre-1986), 4(2), 125.
- Energy Information Administration - Eia - Independent Statistics and Analysis.” Petroleum, Natural Gas, and Coal Still Dominate U.S. Retrieved May 13, 2021
- Fattouh, B., & Mahadeva, L. (2013). OPEC: what difference has it made?. Annu. Rev. Resour. Econ., 5(1), 427-443.
- Fattouh, B. (2021). Saudi Oil Policy: Continuity and Change in the Era of the Energy Transition. Oxford Institute for Energy Studies.
- Fraiberger, S. P., Lee, D., Puy, D., & Ranciere, R. (2018). Media sentiment and international asset prices (No. w25353). National Bureau of Economic Research.
- Hyndman, K. (2008). Disagreement in bargaining: An empirical analysis of OPEC. International Journal of Industrial Organization, 26(3), 811-828.
- Guidi, M. G., Russell, A., & Tarbert, H. (2006). The effect of OPEC policy decisions on oil and stock prices. OPEC review, 30(1), 1-18.
- Kydland, F. E., & Prescott, E. C. (1977). Rules rather than discretion: The inconsistency of optimal plans. Journal of political economy, 85(3), 473-491.
- Lin, S. X., & Tamvakis, M. (2010). OPEC announcements and their effects on crude oil prices. Energy Policy, 38(2), 1010-1016.
- MacKinlay, A. C. (1997). Event studies in economics and finance. Journal of economic literature, 35(1), 13-39.
- Nakov, A., & Pescatori, A. (2010). Monetary policy trade‐offs with a dominant oil producer. Journal of Money, Credit and Banking, 42(1), 1-32.
- Nazer, Y., and Love. A. (2020). The Impacts of New Technology on Market Power: The Case of the US Petroleum Industry. Paper presented at the Mathematical Institute of Oxford University.
- Pevzner, M., Xie, F., & Xin, X. (2015). When firms talk, do investors listen? The role of trust in stock market reactions to corporate earnings announcements. Journal of Financial Economics, 117(1), 190-223.
- Rich, R. W., Raymond, J. E., & Butler, J. S. (1992). The relationship between forecast dispersion and forecast uncertainty: Evidence from a survey data—arch model. Journal of Applied Econometrics, 7(2), 131-148.
- Rösch, A., & Schmidbauer, H. (2009, July). Action Selection in Customer Value Optimization: An Approach Based on Covariate-Dependent Markov Decision Processes. In DMIN (pp. 189-192).
- Ulatowski, R. (2020). OPEC+ as a new governor in Global Energy Governance. Revista UNISCI, (53).
- Summer, D. A., & Mueller, R. A. (1989). Are harvest forecasts news? USDA announcements and futures market reactions. American Journal of Agricultural Economics, 71(1), 1-8.
- Zhang, S., Andrews-Speed, P., Zhao, X., & He, Y. (2013). Interactions between renewable energy policy and renewable energy industrial policy: A critical analysis of China's policy approach to renewable energies. Energy Policy, 62, 342-353.

### Data Appendix — data sources and construction
- OPEC members crude oil production and allocation data
  - Crude oil production data source: IEA MODS Platform.
  - Production data ranged from 1984 to 2020.
  - OPEC organization provides oil production data only from 2001.
  - U.S. Energy Information Administration (EIA) reports OPEC crude oil production data in quarterly, monthly, and annual forms, but only since 1994.
  - OPEC allocation data are taken from OPEC bulletin publications: 1999, 2005, and 2020.

- Oil price data
  - Brent crude oil price data: 1985 to 2019 obtained from FRED.
  - "Three-month" Brent futures contracts data: 1988 to 2019 obtained from Bloomberg terminal DataStream.
  - Brent chosen as the European benchmark price to minimize regional bias relative to WTI and Dubai.

- Macroeconomic data and composite variables
  - World Economic Outlook (WEO) database used for macroeconomic data.
  - I. GDP forecast and GDP forecast standard deviation
    - Arithmetic weighted average indices constructed for real-world GDP forecasts and GDP standard deviation from 1989 to 2019 using the IMF consensus forecast database.
    - The index is weighted based on the current percentage change of real GDP and next year's forecast.
    - The arithmetic weighted average of the GDP standard deviation index is derived using the same method.
    - The real GDP forecast index used as a proxy for uncertainty; the real GDP standard deviation forecast index used as a confidence interval bound for economic downfall measurements.
  - II. Global crude oil production and stock shares
    - IEA MODS Platform database of crude oil production used from 1989 to 2019 to calculate OPEC market share: Total OPEC share, OPEC GCC market share (Saudi, UAE, Kuwait) and Saudi Arabia market share.
    - Market share calculated by dividing each sub-group by global crude oil production.
    - Total oil stock of OECD countries from 1989 to 2019 used as a measure of OPEC policy response to changes in global oil stock.
  - III. Macroeconomic variables
    - Several macroeconomic variables obtained from FRED used in multinomial logic approach.
    - To capture monetary policy effects on OPEC decisions: three-month U.S. Treasury bills and AAA Moody's Corporate Bond Yield are used.
    - A trade-weighted index of major currencies and goods included to measure the impact of the U.S. exchange rate on oil trade.

### Text analysis: approach and implementation details
- Document representation
  - Python Bag of Words approach used to estimate semantic similarity between documents.
  - Created M×N word count matrix where M is collection of OPEC statements and N is list of words contained in the collection.
  - Each row corresponds to a single OPEC statement; each column corresponds to one of the N unique words.
- Similarity measure
  - Similarity between two documents defined as the cosine angle between two row vectors:
    - Similarity = cos(θ) = A ⋅ B / (||A|| ⋅ ||B||) = (∑_{i=1}^n A_i ⋅ B_i) / ( sqrt(∑_{i=1}^n A_i^2) ⋅ sqrt(∑_{i=1}^n B_i^2) )
    - where n is the number of unique terms; A and B represent two document vectors; A_i and B_i represent the number of times that word i occurs in document A and B, respectively.
- Text preparation steps
  - Remove stop words including pronouns, articles, conjunctions, dates, numbers, etc.
  - Stem all words to their root forms (e.g., agreed, agreeing, agreeable → agree).
  - Apply term frequency-inverse document frequency (TF-IDF) weighting to the term-document matrix to lower weight of terms occurring in many documents.
  - Assign a weight for most frequent words used in statements by applying TF-IDF to capture each word's contribution (Fraiberger, Lee, Puy & Rancier 2018).
- TF-IDF weighting formula used (notation preserved)
  - w_{dj} = log( M / N_d + 1 )
    - w_{dj} is frequency weight for each selected word,
    - M is the number of OPEC statements,
    - N_d is the number of articles in which word i is present in j statement.
  - Adding "1" to the IDF ensures terms with zero IDF are not entirely ignored.

### OPEC events, meetings and event study summaries (1987–2019)
- Table of all OPEC events between period 1987 to 2019 includes:
  - OPEC intentions: cut / boost / neutral.
  - Brent oil average price return before and after announcement date and the average impact of meetings on the market.
  - Note: Average price return is based on natural log.
- Event counts and meeting types
  - Total meetings: 101 meetings.
  - Non-regular (extraordinary) meetings: 30.
  - Regular OPEC meetings usually last two days.
- Selected data points and examples from event tables (exact values preserved)
  - Example single-event entries (Date — Decision — Average Before Meeting — Average After Meeting — Average Price Change):
    - 1987-06-29 — Boost — -0.013 — 0.017 — 0.029
    - 2003-07-31 — Neutral — -0.016 — 0.063 — 0.078
    - 2016-11-30 — (3-days Cumulative Return) 9.33% — (11-days Cumulative Return) 9.15%
    - 2000-09-11 — (3-days Cumulative Return) -14.50% — (11-days Cumulative Return) -13.48%
    - 1990-07-27 — (3-days Cumulative Return) 1.80% — (11-days Cumulative Return) 21.98%
  - Aggregate summary statistics by decision type and decade (excerpted, exact numeric columns preserved)
    - Cut All: Number of Events 240 — Average 0.013 — Max 0.134 — Min -0.104 — Price Return Before Announcement 0.000 — After Announcement 0.095 — Difference -0.150 — Volatility -0.013 — etc.
    - Boost All: Number of Events 19 — Average -0.019 — Max 0.060 — Min -0.131 — Price Return Before Announcement 0.013 — After Announcement 0.220 — Difference -0.135 — Volatility 0.032 — etc.
    - Neutral all: Number of Events 580 — Average 0.003 — Max 0.158 — Min -0.091 — Price Return Before Announcement -0.004 — After Announcement 0.107 — Difference -0.161 — Volatility -0.007 — etc.
    - All all: Number of Events 1010 — Average 0.001 — Max 0.158 — Min -0.131 — Price Return Before Announcement 0.000 — After Announcement 0.220 — Difference -0.161 — Volatility -0.001 — etc.
- Notes on event interpretation
  - R denotes oil price cumulative return in the [3] and [11] days after the conclusion of the meeting.
  - If the decision was unexpected: a cut implies R>0, a boost R<0, and no change R=0.
  - Meetings are split into regular (ordinary) and non-regular (extraordinary); non-regular usually called in response to exceptional circumstances (e.g., after the 9/11 terrorist attack).

*Italic: Source — wpiea2022183-print-pdf (References and Data Appendix) — IMF staff material as provided in the content unit.*

---


_Source: https://www.imf.org/-/media/files/publications/wp/2022/english/wpiea2022183-print-pdf.pdf_