Introduction
The relationship between job market sentiment and actual unemployment has long intrigued economists and data scientists. In a recent analysis published on Habr, an independent researcher examined whether the hh.Index—a proprietary metric based on user behavior on the Russian job platform HeadHunter (hh.ru)—can serve as a reliable leading indicator for unemployment rates. The study spans seven years of monthly data, from January 2019 to December 2025, comparing the hh.Index against official Rosstat unemployment statistics. The findings challenge conventional wisdom and offer practical insights for labor market forecasting.
What Is the hh.Index?
The hh.Index is a composite metric calculated by HeadHunter based on the ratio of active resumes to open vacancies on its platform. It is defined as:
hh.Index = (Number of Active Resumes) / (Number of Open Vacancies)
A higher index indicates more job seekers per vacancy, suggesting a tighter labor market from the employer's perspective. The index is normalized and published monthly, making it a real-time proxy for supply-demand dynamics in the Russian labor market. Unlike official unemployment statistics, which are released with a lag of several weeks, the hh.Index is available almost immediately—within the first week of each month.
The Seven-Year Dataset
To test the predictive power of the hh.Index, the researcher compiled a dataset covering 84 months (2019–2025). The dataset includes:
- hh.Index values: Monthly averages from HeadHunter's public reports and API.
- Unemployment rate: Monthly data from Rosstat (Federal State Statistics Service), seasonally adjusted.
- Control variables: GDP growth, inflation (CPI), and labor force participation rate (also from Rosstat).
| Year | Average hh.Index | Average Unemployment Rate (%) |
|---|---|---|
| 2019 | 5.2 | 4.6 |
| 2020 | 6.8 | 5.8 |
| 2021 | 5.5 | 4.8 |
| 2022 | 4.9 | 4.0 |
| 2023 | 5.1 | 3.8 |
| 2024 | 5.4 | 3.5 |
| 2025 | 5.0 | 3.2 |
Data source: HeadHunter public reports; Rosstat monthly bulletins.
The table shows a clear inverse correlation: as the hh.Index declined from 2020 to 2025, unemployment also fell. However, correlation does not imply causation or predictive power—a formal statistical test is required.
Methodology: Granger Causality Test
The researcher applied a Granger causality test to determine whether past values of the hh.Index can predict future unemployment rates. The test was run with lags of 1, 3, and 6 months to account for different time horizons. The null hypothesis (H0) was: hh.Index does not Granger-cause unemployment rate.
Results
| Lag (months) | F-statistic | p-value | Reject H0? |
|---|---|---|---|
| 1 | 4.21 | 0.043 | Yes |
| 3 | 3.87 | 0.051 | No (marginally) |
| 6 | 2.14 | 0.082 | No |
Significance level: α = 0.05.
The test indicates that the hh.Index has statistically significant predictive power for unemployment only at a one-month horizon. For longer lead times (3 and 6 months), the relationship weakens and loses statistical significance. This suggests that the hh.Index is a short-term leading indicator, not a long-term forecasting tool.
Practical Implications
For economists and HR analysts, the findings have several practical takeaways:
- Real-time monitoring: The hh.Index can complement official statistics by providing a near-real-time pulse of labor market tightness. A sudden spike in the index (e.g., a 10% increase month-over-month) may signal rising unemployment within 4–6 weeks.
- Short-term forecasting: Companies using the hh.Index for workforce planning should focus on one-month predictions. For example, if the index rises from 5.0 to 6.0 in April, a moderate increase in unemployment can be expected in May.
- Limitations: The index does not capture structural shifts—such as demographic changes or skill mismatches—that affect long-term unemployment. It also ignores informal employment, which is significant in Russia.
Real-World Case: COVID-19 Pandemic
The most dramatic test of the hh.Index occurred during the COVID-19 pandemic. In April 2020, the index spiked to 8.5 (the highest in the dataset), reflecting massive layoffs and hiring freezes. Unemployment, which had been at 4.5% in March, rose to 5.9% in May 2020—a lag of exactly one month. This aligns with the Granger test results: the index predicted the unemployment peak with a one-month lead.
Conversely, during the recovery phase (2021–2022), the index fell steadily, and unemployment followed suit with a similar short lag. This consistency reinforces the index's utility as a short-term indicator.
Limitations of the Study
The researcher acknowledges several limitations:
- Data granularity: The analysis uses monthly averages, which may obscure weekly or daily fluctuations.
- Platform bias: HeadHunter primarily covers white-collar and formal-sector jobs. Blue-collar and informal workers are underrepresented.
- External shocks: Events like the 2022 sanctions and mobilization altered labor market dynamics in ways not captured by the index alone.
- Stationarity: The time series were differenced to achieve stationarity, which may reduce the interpretability of results.
Comparison with Alternative Indicators
How does the hh.Index stack up against other leading indicators?
| Indicator | Type | Lead Time | Data Availability |
|---|---|---|---|
| hh.Index | Job market sentiment | 1 month | Real-time (weekly) |
| PMI (Manufacturing) | Business activity | 2–3 months | Monthly |
| Consumer Confidence Index | Consumer sentiment | 3–6 months | Monthly |
| Initial Jobless Claims | Unemployment filings | 1–2 weeks | Weekly (U.S.-specific) |
The hh.Index's main advantage is its speed—it is updated weekly, while official unemployment data lags by 3–4 weeks. However, it lacks the breadth of PMI or consumer confidence surveys, which capture broader economic sentiment.
Conclusion
The seven-year analysis shows that the hh.Index can predict unemployment with a one-month lead, though its predictive power diminishes beyond that horizon. For practitioners, this means the index is a valuable short-term tool—especially during periods of rapid change—but should not be relied upon for long-term strategic decisions. The COVID-19 case study validates the index's real-world utility, while the Granger causality test provides statistical rigor.
For those interested in integrating such labor market indicators into their analytics pipeline, ASI Biont supports connecting to HeadHunter's API for automated data retrieval—details are available at asibiont.com/courses.
The study's full dataset and code are available in the original Habr article. As with any single indicator, the hh.Index is best used in combination with other metrics for robust forecasting.
Comments