左栏为原创英文研读稿和中文译写,右栏批注与研读单元对应。研读单元按作者公开工作论文的章节编排;需要比对原文段落时,可按节打开作者稿对应页。
图表索引 · 62 项
- Table 1: Intangibles selling industries
- Table 2: Intangible investments
- Table 3: Intangibles by sector and firm size class
- Figure 1: Intangible investment per worker shares around events
- Table 4: Firm-level outcomes for intangibles investing vs non-intangible investing firms
- Table 5: Production function estimation
- Figure 2: Effect of intangible investment spike on productivity mismeasurement
- Table 6: Effect of intangible investment spikes on productivity mismeasurement
- Figure 3: Placebo events: Spikes in material fixed assets
- Figure 4: Effect heterogeneity
- Table 7: Effect heterogeneity
- Table 8: Other firm-level outcomes
- Figure 5: A tangible example: ICT and the productivity slowdown
- Figure A.1: Example of a VAT declaration form
- Figure A.2: Example of a customer listings form
- Table A.1: Intangibles selling industries
- Figure A.3: The evolution of intangible investment type by data source
- Figure A.4: Intangible intensity in manufacturing and services industries
- Table A.2: Summary statistics for the steps of the construction of the final sample of firms
- Table A.3: Descriptive statistics
- Figure A.5: The distribution of the main variables
- Figure A.6: The distribution of the intangible investment types
- Figure A.7: The importance of tangible and intangible investment over time
- Figure A.8: Intangible capital deepening
- Figure A.9: Tangible capital deepening
- Figure A.10: Comparison with sectoral intangibles data from EUKlems
- Table C.1: Intangible spike frequency
- Table C.2: Share of firms ever having an intangible investment spike
- Figure C.1: Intangibles expenditures and total capital around intangible investment spikes
- Table C.3: Correlates of a firm ever having an intangible investment spike
- Table D.1: Correlates of a firm ever having an intangible investment spike
- Figure D.1: Average intangible-to-tangible capital for firms with and without intangible investment
- Figure D.2: Cross-sectional relationship between intangibles per employee and firm size
- Table D.2: Regressions of log intangibles per employee on log firm size
- Table D.3: Intangible investments from the balance sheets
- Table D.4: B2B-based intangible investments and balance sheet intangibles by sector
- Table D.5: B2B-based intangible investments and balance sheet intangibles by firm size class
- Table D.6: Intangible investments without fees
- Table D.7: B2B-based intangible investments and B2B without fees by sector
- Table D.8: B2B-based intangible investments and B2B without fees by firm size class
- Table D.9: Intangible investments with imports
- Table D.10: B2B-based intangible investments and B2B with imports by sector
- Table D.11: B2B-based intangible investments and B2B with imports by firm size class
- Table D.12: Production function estimation
- Table E.1: Intangible investments
- Table E.2: Intangible spike frequency by type
- Table E.3: Intangibles by sector and firm size class
- Figure E.1: Computerized intangible investment shares around events
- Figure E.2: Innovative intangible investment shares around events
- Figure E.3: Economic competencies intangible investment shares around events
- Figure F.1: Robustness to different definitions of intangible investment events
- Table F.1: Tangible investments
- Table F.2: Tangible spike frequency
- Figure F.2: Robustness to outliers
- Table F.3: Intangible spike frequency by type
- Figure F.3: The impact of different measures of intangible capital on productivity mismeasurement
- Figure F.4: Using later-investing vs non-investing firms as control group
- Figure F.5: Robustness to matching on firm size and/or age
- Figure F.6: Using output elasticities of production function without intangibles
- Table F.4: ICT selling industries
- Figure F.7: AI usage within firms
- Table F.5: Usage of various AI technologies across companies of different sizes.
摘要 Abstract
Firms can spend heavily on capabilities whose services arrive over several years. When the purchase enters the accounts as a current input, conventional productivity can look weaker even as productive capacity is being built. The paper asks how large that discrepancy is around episodes of unusually high spending.
企业可能集中购入多年后仍能发挥作用的能力。如果会计上把购置支出全计入当期投入,即使企业正在积累生产能力,常规生产率也可能表现疲弱。论文据此追问,在支出异常集中的时期,这一差距究竟有多大。
The study follows Belgian firms around sharp increases in intangible spending and compares a standard measure of total factor productivity with one that accounts for intangible investment. By the end of the seven years beginning with the event, the latter measure is about 3% higher. The divergence is more pronounced among younger, smaller and less capital-intensive firms.
研究跟踪比利时企业无形支出陡增前后的变化,对比常规全要素生产率与计入无形投资的估计。自事件发生当年起算七年,期末调整口径比常规口径约高 3%。这一差距在较年轻、较小和资本密集度较低的企业中更为明显。
1. 引言 Introduction
The introduction begins with the tension between expanding digital capabilities and sluggish measured productivity. The productivity slowdown is the broad puzzle. The tractable question is whether firms that acquire software, knowledge and organizational capabilities are producing assets that the usual accounting boundary misses.
引言从数字能力扩张与生产率统计表现疲弱之间的张力切入。生产率增速放缓是宏观谜题。作者进一步把问题收窄,考察企业购买软件、知识和组织能力所形成的资产,有多少被常用核算边界排除在外。
Aggregate studies can suggest the scale of omitted investment, and detailed case studies can show how a specific asset is used. Neither level alone reveals how the discrepancy evolves within many different firms. The paper fills this gap by following the timing of spending and comparing responses across firms.
总量研究可以估计被漏计的投资规模,细致的案例则能说明某种资产如何使用。要观察不同类型企业内部的误测随时间怎么演变,需要能够跨企业、跨期比较的微观记录。本文因此利用投资发生的时点,并比较企业之间的反应差异。
The authors use the seller's activity to classify purchases in Belgian business-to-business transaction records. By tracing these purchases repeatedly for the same buyer, they extend observation beyond firms that capitalize intangibles on their balance sheets and give the main hypothesis a measurable input.
作者根据卖方的经营活动,对比利时企业间交易记录中的采购进行分类。连续追踪同一买方的采购,使研究能够覆盖账上未确认无形资产的企业,也让核心假设所需的无形投入有了可观察指标。
The empirical argument combines two comparisons. First the paper constructs a standard and an intangible-adjusted productivity series for the same firms. Then it asks whether their difference shifts at the time of a spending spike, relative to the contemporaneous change in firms without a spike in the same observation window.
实证论证由两重比较组成。作者先为同一企业构造常规生产率与无形投资调整后的生产率,再检验两者之差是否在集中支出时点发生变化,并以同一观察窗口内未出现峰值的企业作为参照。
2. 理论框架 Theoretical framework
The theory section defines two comparable measurement rules before looking for a dynamic pattern in the data. It moves from aggregate accounting to firm-level estimation.
理论节先界定两种可对照的测算规则,再讨论数据中可能看到的变化轨迹。论证从总量核算转到企业层面的估计。
2.1 总量核算框架 Aggregate accounting framework
The aggregate framework distinguishes investment that creates an asset from purchases whose services are used up in the period. Once intangible asset creation is counted as part of output, the Solow residual must also reflect the services of the accumulated intangible stock. Both the output boundary and the input list therefore change.
总量框架区分形成资产的投资与当期耗尽的采购。一旦把无形资产的形成计入产出,计算索洛余值时也要扣除已积累无形资本所提供的服务。因此,产出边界与投入项目都需要调整。
When spending surges, conventional statistics fail to recognize some asset creation in that period. Once an intangible stock is in use, the omitted capital services can instead be credited to the residual in the conventional calculation. The relative strength of these forces can shift as firms move from investment to use.
支出集中增长时,常规统计漏记了一部分当期新形成的资产。等存量无形资产进入使用阶段,常规测算又可能把漏计的资本服务归入生产率余值。企业从集中投入转向使用资产,两种力量的相对大小便可能发生变化。
2.2 企业层面模型 Firm-level model
The firm-level model brings in intermediate inputs alongside labor and tangible capital. The authors contrast an accounting treatment that leaves qualifying purchases among intermediates with an adjustment that removes those purchases from current intermediates and introduces an intangible capital stock.
企业层面的模型在劳动和有形资本之外还包括中间投入。作者把两种处理放在一起比较。一种把符合条件的采购留在当期中间投入中,另一种把这些支出扣除,并把由此形成的无形资本存量纳入生产函数。
The paper uses the perpetual inventory method to build real intangible capital. Current investment adds to the remaining stock after depreciation. The resulting series lets the authors ask how a one-year spending surge can leave a stock that contributes to production over several subsequent years.
论文采用永续盘存法构造实际无形资本。前期存量扣除折旧,再加本期实际投资。由此得到的序列可以用来观察,一年的集中支出如何留下在后续几年持续参与生产的资本存量。
The formulas identify two channels, but do not determine their relative size at each date. A spending spike should first strengthen the current-input channel. The later trajectory depends on how fast the stock grows and how much productive service the firm obtains from it.
公式给出了两条误测渠道,但无法确定它们在每个时期各有多大。发生投资峰值时,支出记入当期中间投入的影响预计先增强。后续路径则取决于无形资本积累的速度,以及企业能从中获得多少生产服务。
3. 数据 Data
This section translates an accounting distinction into observable measures of a firm's intangible purchases and checks their distribution and coverage.
数据节把费用与资本的核算区分转为可观测的企业无形采购指标,并检查这些指标的分布与覆盖。
3.1 数据来源与测量 Data sources and measurement
The firm panel covers Belgium's nonfinancial private sector over 2002–2019. Purchase records identify spending on externally supplied intangibles. The data construction also draws on firm records and adds information on in-house research and development where available.
企业面板覆盖 2002 至 2019 年比利时非金融私营部门。采购记录用于识别外部提供的无形投入。数据构建还使用企业资料,并在可得时补充企业内部研发信息。
The classification separates computerized information, innovative property and economic competencies. Software, research or design, and training or organizational capabilities can accordingly enter different components. The distinction makes it possible to test whether the result is concentrated in a particular family of assets.
分类将无形资产分为计算机化信息、创新资产和经济能力。软件、研发或设计,以及培训或组织能力等支出相应归入不同类别。这个区分让作者能进一步检验,主结果是否主要来自某一类资产。
Supplier-based classification makes comparable measurement possible across many buyers, but small purchases below the reporting threshold and many internally generated capabilities remain hard to observe. A missing transaction has a different evidentiary meaning from a verified absence of investment.
基于供应商的分类使大量买方可以在同一口径下比较。不过,低于申报门槛的采购和许多企业自创的能力仍难观察。没有出现在交易记录中,只能说明该指标未观测到这类投资。
3.2 描述统计 Descriptive statistics
At the firm-year level, intangible spending is strongly right-skewed. The working paper reports an average spending-to-capital ratio of 6.38% and a median of about 2%. Larger firms also show higher spending per worker, and their spending is larger relative to capital in the reported size breakdown.
企业—年份层面的无形支出呈明显右偏。作者稿报告的支出占总资本比率平均为 6.38%,中位数约为 2%。按规模分组,较大企业的人均支出更高,支出占资本的比例也更高。
3.3 与资产负债表口径对照 Comparison with balance-sheet intangibles
The balance-sheet comparison shows why book values are an incomplete substitute for spending records. The working paper reports zeros for 82.7% of balance-sheet firm-years versus 18.6% under the transaction-based measure in its comparison. Recognition rules and differing disclosure practices help explain why the two sources show different coverage.
作者用资产负债表数据检验账面指标能在多大程度上替代采购记录。在作者稿的对照中,资产负债表口径下企业—年份为零的比例是 82.7%,交易口径则为 18.6%。资产确认规则和披露差异有助于解释两种来源为何覆盖不同。
4. 无形投资事件 Intangible investment events
The paper defines an observable investment event, then examines which firms and years satisfy the rule.
论文首先定义可观察的投资事件,然后检查哪些企业和年份满足判定规则。
4.1 投资峰值的定义 Defining investment spikes
An event comes from a firm-specific comparison. In the baseline, the current real intangible spending share must be at least three times its average in the firm's other observed years. The first year meeting that rule marks the event in the analysis.
事件通过企业自身的历史水平来识别。在基准口径中,当年实际无形投资占比须至少达到该企业其余观测年份平均水平的三倍。首次达到这一条件的年份被定为事件年。
The authors treat the spike as an indicator of unusually concentrated spending. They avoid making the spike's reported euro amount the main treatment dose, given uncertainty about in-house creation and variation in the assets purchased. A firm without a qualifying event can still invest in intangibles steadily.
作者把峰值用作支出异常集中的指标。考虑到企业内部形成的能力难以完整测量,购入资产的构成也有差异,作者采用是否发生峰值作为主要处理口径。企业即便从未跨过门槛,也可能一直在进行无形投资。
4.2 投资事件的描述统计 Descriptive evidence on investment events
At least one spike is observed for 48.9% of firms. Qualifying years make up 6.7% of firm-year observations, and event-time plots show a concentrated increase in expenditure per worker. Thus, spikes are present in many firms even though they occur in a small fraction of observed years.
有 48.9% 的企业至少出现过一次峰值。满足峰值定义的企业—年份占全部观察的 6.7%,事件时间图也显示人均支出在峰值年集中上升。因此,经历过峰值的企业相当多,峰值本身却集中在少数观测年份。
Figure 1: Intangible investment per worker shares around events
无形投资峰值前后的人均投入
5. 实证方法 Methodology
The authors begin with observable differences between firms and then build comparisons aligned around the first investment spike.
作者先展示企业之间可观察的差异,再以首次投资峰值对齐企业进行比较。
5.1 比较企业的变化轨迹 Comparing firm trajectories
Firms that experience a spike look different from firms that do not. Descriptive regressions associate spike status with faster growth in employment, value added and the intangible-to-tangible capital ratio. The differences make an unconditional comparison hard to attribute to investment timing alone.
发生过峰值的企业与没有发生过峰值的企业表现不同。描述性回归发现,前一类企业的就业、增加值以及无形资本与有形资本之比增长更快。这些差异表明,直接比较两类企业,很难把差距归因于投资时点。
Table 4: Firm-level outcomes for intangibles investing vs non-intangible investing firms
无形投资企业与其他企业的经营结果
5.2 识别策略 Identification strategy
The stacked difference-in-differences design builds one comparison dataset for each first-event cohort from 2005 to 2012. It aligns firms from three years before the event through six years after it. For each cohort, the control firms experience their first spike only after that cohort's window closes.
堆叠双重差分按照 2005 至 2012 年的首次事件年份建立不同队列。每个队列以事件为中心,覆盖前三年至后六年。该队列的对照企业虽也会发生首次峰值,但时间在它的观察窗口结束以后。
The regression measures relative changes from the year immediately before the first spike. It absorbs persistent differences with firm-by-cohort fixed effects and common sector shocks with industry-by-calendar-year fixed effects. A given firm can appear in more than one cohort dataset.
回归以首次峰值前一年为参照,估计各事件时点的相对变化。企业与队列的固定效应吸收长期差异,行业与日历年份的固定效应吸收行业共同冲击。同一企业可能出现在多个队列数据集中。
The causal interpretation requires parallel trends in the absence of the event and no anticipatory change in the outcome. The two coefficients for the years before the reference year offer a diagnostic. An apparently flat pre-event path provides only limited evidence about unobserved shocks at the event.
因果解释要求,如果未发生投资事件,处理组与对照组的走势应保持平行,而且结果变量不会因预期投资而提前变化。基准年前两个时点的系数可用来诊断此前走势。即便事件前看起来平稳,对事件当年不可观察冲击的排除仍然有限。
6. 无形投资会引起生产率 J 曲线吗? Do intangible investments cause a Productivity J-curve?
After estimating production inputs, the article compares two productivity paths around the same event and checks alternatives through a placebo, subgroup results, and robustness exercises.
论文先估计各生产投入的贡献,再比较同一事件周围的两套生产率路径,并用安慰剂、分组结果和稳健性检验考察替代解释。
6.1 生产函数 Production functions
Firms respond to their productivity when choosing inputs, and capital stocks are measured with error. The paper uses a two-step control-function procedure with lagged investment instruments for capital to address these problems before constructing the two productivity series.
企业会根据自身生产率决定投入,资本存量本身也存在测量误差。作者在计算两套生产率序列前,用两步控制函数方法,并用滞后投资作为资本投入的工具变量,应对这两类问题。
The working paper reports an estimated intangible-capital output elasticity of roughly 0.01 to 0.06 across specifications. It then scales elasticities by output-to-input ratios to compare marginal products across factors. The implied value for intangible capital is about €0.98 of sales per additional euro of the capital input in the reported calculation.
作者稿在不同设定下估计,无形资本的产出弹性约为 0.01 至 0.06。作者再以产出与投入的比值对弹性作转换,比较不同要素的边际产出。相关计算得到,每多一欧元的无形资本投入对应约 0.98 欧元的销售额边际变化。
6.2 双重差分结果 Difference-in-differences results
Figure 2 follows the growth-rate difference between conventional and adjusted productivity. The divergence is about 0.80 percentage points in the event year and 0.89 in the following year in the working paper's log-growth units. The estimated annual gap is much smaller by the end of the window.
图 2 跟踪常规口径与调整口径生产率的增长率差。按作者稿的对数增长单位,常规口径在事件年约低 0.80 个百分点,在随后一年约低 0.89 个百分点。到观察窗末端,估计的年度差距已明显收窄。
Summing the event-year and six subsequent annual differences produces an end-of-window gap of about 3% in productivity levels. A small annual gap at the end can therefore coexist with a sizable gap accumulated over the period.
把事件年及随后六年的年度差逐项累加,得到窗口终点约 3% 的生产率水平差。到末期,年度增长率差可能很小,而此前积累的生产率水平差依然可观。
The expenditure component becomes important immediately when the firm records a spending spike. The stock component adjusts more gradually as capabilities accumulate. To examine different assets more closely, the authors also estimate category-specific events after excluding windows with spikes in competing intangible categories.
企业出现集中支出时,支出计入中间投入的渠道立即变得重要。资本存量渠道则随着能力的积累逐步改变。为了分辨不同资产类别的作用,作者还估计只涉及单类资产的事件,并排除窗口内其他无形资产类别也发生峰值的观测。
Figure 2: Effect of intangible investment spike on productivity mismeasurement
无形投资峰值对生产率误测的影响
Table 6: Effect of intangible investment spikes on productivity mismeasurement
按事件期和投资类别估计生产率误测
6.3 有形投资安慰剂 Tangible-investment placebo
The placebo reruns the event design around spikes in material fixed assets. These investments also rise sharply in the event year. Figure 3, however, shows no discernible shift in the intangible-adjustment gap around the tangible spike. The comparison tests whether any large outlay would mechanically produce the observed profile.
安慰剂检验把同一事件设计用于有形固定资产投资峰值。有形投资同样在事件年陡增,但图 3 显示,无形调整造成的口径差在有形投资峰值附近没有明显变化。这个对照检验了任何大额支出是否都会机械地产生相同曲线。
6.4 企业异质性 Differences across firms
The event-time profiles are more pronounced for younger, smaller and less capital-intensive firms. Plotting separate trajectories reveals differences that a pooled estimate would conceal. The authors link these dimensions to variation in firms' existing capabilities and their ability to absorb new investments.
事件时间曲线在较年轻、较小和资本密集度较低的企业中表现更明显。分组画出各自路径,可以看出总体估计掩盖的差异。作者把这些差异与企业原有能力和吸收新投资的能力联系起来。
6.5 异质性解释 Explaining the differences
The paper studies other outcomes, including changes in employment, value added and the intangible capital mix. These outcomes are used to assess whether absorptive capacity and the need for complementary investments can explain a more pronounced adjustment among firms that begin with fewer resources.
作者进一步考察就业、增加值和资本构成等结果,判断吸收能力的差异以及补充投资需求,能否解释初始资源较少的企业何以表现出更明显的调整。
6.6 稳健性检验 Robustness checks
The authors vary the baseline spike threshold from two to five times the firm's usual share, try alternative denominators and test the influence of extreme observations. They also consider balance-sheet data in place of transaction-based intangible spending and recompute the comparison using alternative production elasticities.
作者把投资峰值门槛改为企业通常投资占比的二至五倍,替换分母,并检查极端观测值是否左右结果。他们还用资产负债表数据替代基于交易记录的无形投入指标,并用另一套生产弹性重算口径差。
The paper also compares treated firms with firms that never record a spike in the observation period and with nearest neighbors matched on pre-event characteristics. Similar patterns under these comparisons reduce concern about one particular control construction. They leave open selection on unmeasured, time-varying conditions.
论文还用观测期内从未出现峰值的企业作对照,并根据事件前特征进行最近邻匹配。如果更换对照方式后仍出现类似路径,就能降低结果完全由某种对照组设定所致的疑虑。随时间变化的不可观测因素仍可能影响选择。
7. 信息通信技术、人工智能与生产率放缓 ICT technologies, artificial intelligence, and the productivity slowdown
Figure 5 extends the event logic to investment in information and communication technology. It contrasts ICT spikes accompanied by intangible spikes with ICT-only comparisons. The point is to see whether the accounting gap is concentrated in episodes where technology is combined with complementary intangible capabilities.
图 5 把事件研究的思路用于信息通信技术投资,对照同时伴随无形投资峰值和仅发生信息通信技术投资峰值的情形。作者由此考察,当技术设备投入与配套的无形能力建设结合时,生产率测量偏差是否更集中地出现。
The paper considers AI both as a tool for improving specific tasks and as a technology that can change how firms innovate. The second route requires complementary investment in workers, data and work processes. Those outlays could make measured performance improve later than the underlying capability.
作者讨论人工智能的两类作用,一是改善具体任务,二是改变企业的创新方式。后一类作用需要劳动力、数据和工作流程等配套投资。正是这些支出,可能使常规绩效统计晚于生产能力本身的改善。
The authors draw on prior research showing that AI adoption differs across firms. They suggest that a broader rollout among less digitally mature firms could bring adjustment costs into view. The Belgian investment-event results do not estimate a generative-AI treatment effect or an economy-wide contribution.
作者引用既有文献说明,不同企业采用人工智能的程度有别。等数字化基础较弱的企业也开始大范围采用时,调整成本才可能更明显地反映在统计中。比利时企业的投资事件估计无法直接给出生成式人工智能的处理效应或对总体经济的贡献。
Figure 5: A tangible example: ICT and the productivity slowdown
ICT 投资与生产率放缓
8. 结论 Conclusion
The paper concludes that acquiring intangible assets can change the timing recorded by conventional productivity statistics. Recognizing intangible investment changes the estimated path, especially for smaller, younger and less capital-intensive firms. This makes the accounting boundary consequential for evaluating firm performance.
论文认为,购入无形资产会改变常规生产率统计所呈现的时间路径。若在测算中确认无形投资,生产率路径将随之改变,较小、较年轻和资本密集度较低的企业尤其如此。因此,评价企业绩效时,应明确核算边界如何处理无形投入。
An early dip after adoption can combine true transition costs with a measurement effect. The accounting results give a reason to inspect the components before interpreting a weak productivity reading. The relative importance of these forces outside the studied firms remains an empirical question.
采用新技术后的早期回落可能同时包含真实的转型成本和测量因素。论文的核算比较提示,面对疲弱的生产率数据,应先分辨这两部分。在本文所研究企业以外,两种因素各自有多重要,仍有待新的证据回答。
参考文献研读 References
The bibliography supports distinct parts of the argument: intangible accounting, productivity estimation, and staggered event studies. Following each line of work clarifies which ideas the authors borrow and which they test with Belgian firm data.
参考文献分别支撑无形投入核算、生产率估计与交错事件研究设计。从这三条线索追读,可以区分论文沿用的框架与用比利时企业数据检验的命题。
附录研读索引 Appendices
Appendix A documents how purchases enter the intangible asset taxonomy and how stocks are constructed. Appendix B gives additional detail on production-function estimation. Read together, they provide the practical route from raw expenditure records to the productivity series used in the event study.
附录 A 说明如何将采购归入不同无形资产类别并构造资本存量。附录 B 补充生产函数的估计细节。两节连起来看,可以追溯从原始支出记录到事件研究所用生产率序列的具体步骤。
Appendices C through E add detail on investment events, firm-level comparisons and spikes by asset type. In particular, event-time plots let the reader compare the short-lived spending peak with the slower movement of the accumulated stock.
附录 C 至 E 补充投资事件、企业层面比较和各资产类别的投资峰值。事件时间图尤其便于核对,集中支出在峰值年迅速上升,而累积资本存量的变动较为缓慢。
Appendix F contains supplementary checks for the event design and measurement choices. Appendix G expands the discussion of ICT and complementary intangible investment. The first set probes the stability of the central estimate, while the second explores where the mechanism might also matter.
附录 F 补充事件设计和测量设定的稳健性检验。附录 G 延伸讨论信息通信技术及其配套无形投资。前者帮助判断主估计是否稳定,后者帮助判断同一机制还可能应用于哪些情形。
附录 A 数据构建 A Data and measurement
The appendix traces how raw records become the inputs for the productivity model.
附录 A 追溯原始记录如何转化为生产率模型所用的投入变量。
A.1 行政记录与问卷 Source records
Different administrative and survey datasets supply different margins of firm activity.
多套行政记录与问卷分别提供企业活动的不同侧面。
A.1.1 比利时央行企业账目 Company accounts
The company accounts supply variables such as value added, capital, and employment, with different reporting detail across firms.
企业账目提供增加值、资本和就业等变量,企业之间的填报详细程度有差别。
A.1.2 增值税申报 VAT returns
The authors use VAT declarations to observe acquisitions of investment goods and firms’ sales and purchases.
作者用增值税申报资料观察企业购置的投资品,以及销售和采购情况。
A.1.3 企业间交易 Buyer–seller records
Domestic purchases can be reconstructed from customer listings filed by sellers.
根据卖方填报的客户清单,可以反推企业在本国的采购。
A.1.4 研发支出调查 Research spending survey
Survey data provide an additional window on research and development conducted within firms.
研发调查为观察企业内部开展的研发活动提供了另一种途径。
Table A.1: Intangibles selling industries
无形资产供应业详细代码
Figure A.3: The evolution of intangible investment type by data source
按来源拆分的无形投资时间变化
A.2 无形投入的行业强度 Variation across industries
The appendix shows sector-level dispersion separately for manufacturing and services.
附录分别呈现制造业和服务业内部各行业的无形投入强度差异。
Figure A.4: Intangible intensity in manufacturing and services industries
制造业和服务业无形投资强度
A.3 生产函数的样本构建 Estimation sample
The authors join administrative records using firm VAT identifiers, then apply eligibility restrictions before estimating production functions.
作者通过企业增值税编号连接行政记录,再依据样本规则筛选企业后估计生产函数。
Table A.2: Summary statistics for the steps of the construction of the final sample of firms
企业最终样本的构建步骤
Figure A.6: The distribution of the intangible investment types
三类无形投资的分布
A.4 实际无形资本存量 Building real capital stocks
The perpetual inventory calculation links previous stock, depreciation, and new deflated investment.
永续盘存法把上期存量、折旧和扣除价格变化后的新增投资联系起来。
A.5 无形资本深化 Trends over time
The authors chart changes in aggregate real spending over the sample years.
作者绘制样本期内实际无形投入总量的变化路径。
Figure A.7: The importance of tangible and intangible investment over time
有形和无形投资的重要性
A.6 与 EU KLEMS 加总数据对照 External benchmark
The EU KLEMS comparison places Belgian investment intensity beside other countries’ measured trends.
EU KLEMS 对照把比利时投资强度与其他国家测得的走势并列展示。
Figure A.10: Comparison with sectoral intangibles data from EUKlems
与 EUKLEMS 行业无形资产数据比较
附录 B 生产函数估计 B Production-function estimation
The appendix explains how control-function estimators address input choices that respond to unobserved productivity.
附录 B 解释控制函数估计如何应对企业根据不可观测的生产率选择投入这一问题。
附录 C 无形投资事件 C Investment-event diagnostics
Appendix C checks who has an event, what drives it, and how it overlaps with tangible investment.
附录 C 核对哪些企业发生事件、峰值由什么驱动,以及它与有形投资有多少重叠。
C.1 投资峰值频数 How often spikes occur
The frequency table distinguishes firms with no spike, one spike, and repeated spikes.
频数表区分从未出现峰值、只出现一次和重复出现峰值的企业。
C.2 行业与规模组的事件比例 Who experiences a spike
The subgroup table reports how frequently firms ever cross the event threshold across sectors and size classes.
分组表报告不同行业与规模组的企业曾经跨过事件门槛的比例。
Table C.2: Share of firms ever having an intangible investment spike
曾出现无形投资峰值的企业比例
C.3 投资与资本的分别变化 The numerator and denominator
Separate series for spending and capital test whether the investment ratio jumps because of a rise in expenditure.
分别画出支出和资本序列,可检查投资占比的峰值是否由支出上升驱动。
Figure C.1: Intangibles expenditures and total capital around intangible investment spikes
峰值附近无形投入与总资本
C.4 两类投资峰值的关联 Coincidence with tangible investment
The appendix checks whether a firm ever having a tangible spike is also more likely to show an intangible spike.
作者检验曾出现有形投资峰值的企业是否也更可能出现无形投资峰值。
Table C.3: Correlates of a firm ever having an intangible investment spike
出现无形投资峰值的企业特征
附录 D 企业比较与替代测量 D Firm comparisons and measurement
Appendix D links firm characteristics and alternative spending measures to the main analysis.
附录 D 从企业特征和替代支出指标两方面补充正文比较。
D.1 投资事件的预测特征 Predicting events
The appendix relates prior firm characteristics to the probability of ever recording an event.
作者将既有企业特征与观测期间曾有投资事件的概率联系起来。
Table D.1: Correlates of a firm ever having an intangible investment spike
无形投资峰值与企业特征的详细估计
D.2 平衡面板中的资本比率 Capital ratios in a balanced panel
The comparison follows firms observed for the full period and normalizes group paths to the same starting year.
对照限定为整个时期都可观察的企业,并让两组序列以前期同一年为标准进行归一化。
Figure D.1: Average intangible-to-tangible capital for firms with and without intangible investment
有无峰值企业的无形相对有形资本
D.3 企业规模与无形投入需求 Demand and firm size
The size comparison asks whether larger firms have greater intangible spending per worker.
规模比较检验较大的企业是否有更高的人均无形支出。
Figure D.2: Cross-sectional relationship between intangibles per employee and firm size
规模与人均无形投资的横截面关系
Table D.2: Regressions of log intangibles per employee on log firm size
规模与人均无形投资的回归
D.4 替代无形投入指标 Alternative measurement rules
The appendix varies the treatment of book values, low-value fees, and imports before comparing resulting estimates.
作者改变账面值、低额服务费和进口采购的处理方式,再比较产生的估计。
D.4.1 用账面存量反推投资 Reconstructing purchases from book assets
A flow can be reconstructed from book stocks through a depreciation-adjusted change in recorded assets.
通过扣除折旧影响后的账面存量变化,可以反推当期投资流量。
Table D.4: B2B-based intangible investments and balance sheet intangibles by sector
按行业比较交易与资产负债表口径
Table D.5: B2B-based intangible investments and balance sheet intangibles by firm size class
按企业规模比较交易与资产负债表口径
D.4.2 排除可能的低额服务费 Removing small fees
Trimming the bottom of positive purchases probes whether recurring service fees account for the pattern.
截去正支出分布中的低额部分,检验重复性服务费能否解释观察到的形态。
Table D.7: B2B-based intangible investments and B2B without fees by sector
按行业比较扣除费用前后
Table D.8: B2B-based intangible investments and B2B without fees by firm size class
按规模比较扣除费用前后
D.4.3 纳入进口无形投入 Adding imports
The authors supplement domestic transactions with survey information on foreign purchases by firms.
作者用企业境外采购的调查资料补充国内交易记录。
Table D.10: B2B-based intangible investments and B2B with imports by sector
按行业比较是否计入进口
Table D.11: B2B-based intangible investments and B2B with imports by firm size class
按规模比较是否计入进口
D.4.4 替代口径下的生产函数 Re-estimating elasticities
The elasticity table compares the production model after substituting alternative intangible-input measures.
产出弹性表比较更换无形投入指标后的生产函数估计。
附录 E 各类无形投资的峰值 E Asset-type comparisons
The appendix distinguishes spending amounts, the prevalence of spikes, and their event-time paths by asset type.
附录 E 区分不同资产类别的支出金额、峰值普及程度和事件时间走势。
Figure E.1: Computerized intangible investment shares around events
计算机化无形投资峰值路径
Figure E.2: Innovative intangible investment shares around events
创新资产投资峰值路径
Figure E.3: Economic competencies intangible investment shares around events
经济能力投资峰值路径
附录 F 稳健性检验 F Robustness checks
The appendix checks event definitions, comparison firms, input measures, and estimation choices.
附录 F 依次检查事件定义、对照企业、投入计量和估计设定。
F.1 替代事件定义 Event definitions
The authors alter the event denominator and the spike threshold.
作者分别改变确定投资事件的分母和峰值门槛。
Figure F.1: Robustness to different definitions of intangible investment events
替换无形投资事件定义
F.2 有形安慰剂构建 Placebo-event construction
The placebo sample identifies spikes in material fixed assets among firms for which the relevant measures overlap.
安慰剂样本在相关数据重叠的企业中识别有形固定资产投资峰值。
F.3 极端值处理 Extreme observations
The figure compares the baseline with winsorized and trimmed series.
图中将基准估计同缩尾和删除部分极端值后的估计比较。
F.4 不同无形投入指标的事件比较 Comparable investment measures
The comparison keeps an employment-based denominator while varying the intangible measure.
比较使用基于就业规模的分母,同时改变无形投入指标。
Figure F.3: The impact of different measures of intangible capital on productivity mismeasurement
不同无形资本指标下的生产率误测
F.5 从未出现峰值的对照组 Alternative comparison firms
The authors rerun the design with firms without a spike in the observation window as a comparison.
作者以观测期未出现峰值的企业重新运行比较设计。
Figure F.4: Using later-investing vs non-investing firms as control group
使用后期投资企业或从未投资企业作对照
F.6 按企业年龄与规模匹配 Matching on initial conditions
Matching exercises compare results after accounting for pre-event firm size, age, or both.
匹配检验分别按事件前规模、年龄及二者共同匹配后比较估计结果。
F.7 允许两套生产函数的弹性不同 Alternative output elasticities
The authors allow the two productivity calculations to use separately estimated output elasticities.
作者允许两套生产率测算分别采用各自估计的产出弹性。
Figure F.6: Using output elasticities of production function without intangibles
沿用无形资产未入模时的产出弹性
附录 G 技术与互补投资 G Technology and complementary investment
Appendix G develops the link from ICT and AI adoption to the investment-complementarity argument.
附录 G 补充信息通信技术和人工智能采用如何联系到投入互补的论证。
G.1 信息通信技术支出的测量 Measuring ICT purchases
ICT purchases are tracked through transaction records and industry classifications of supplying firms.
信息通信技术采购根据交易记录和供应商行业分类识别。
G.2 ICT 事件的比较方法 The ICT event comparison
The exercise compares ICT events with different degrees of complementary intangible investment.
该分析比较包含不同程度互补性无形投入的 ICT 投资事件。
G.3 人工智能采用调查 AI adoption survey
The Flemish survey documents firm adoption, capabilities, and barriers for AI.
佛兰德斯调查记录企业采用人工智能的程度、相关能力及障碍。
Table F.5: Usage of various AI technologies across companies of different sizes.
按企业规模列示各 AI 技术使用情况
论证结构总览
全文逻辑骨架
- 问题
企业获得软件、研发或组织能力时,常规生产率统计能否识别投入形成的资产?集中投入之后,两种核算口径的差距何时出现、持续多久?
- 论点
集中购入无形投入时,部分形成长期能力的支出被计作当期中间投入。随后无形资本服务逐渐释放,两条核算渠道的相对影响随时间变化。
- 证据
作者用比利时 2002 至 2019 年的企业数据构造两套生产率,按首次无形投资峰值做堆叠双重差分。自事件当年起的七年内,差距先大后小,期末累计水平差约为 3%。
- 反驳与限定
有形投资安慰剂没有呈现类似路径,改变峰值门槛、对照组与匹配方式仍支持主结果。企业同期重组等冲击,以及从微观结果到宏观放缓的外推,仍需要另行验证。
- 结论
核算边界会改变对集中投资后生产率走势的判断,较年轻、较小和资本密集度较低的企业尤其明显。数字化与人工智能的互补投资是后续可检验的应用方向。
作者核心主张
常规口径将形成长期能力的无形投资计作当期投入,因而低估集中投资后数年的企业全要素生产率。
论证最强处
对同一企业构造两套生产率,可以沿投资时点追踪两套口径的分歧;再用同样陡增但通常资本化的有形投资做安慰剂,直接检验通用投资峰值能否解释这一分歧。
论证最弱处
无形投资峰值并非随机发生。若峰值恰与重组、外包或扩张同时发生,这些企业特定冲击可能影响口径差,有限的处理前趋势和匹配无法将其完全排除。