1 Introduction

Policy debates about financial globalisation are closely connected to the investment technology of foreign investors in emerging markets. Do foreign investors bring capital to good projects? Or are foreign investors afflicted with weaknesses of information and analysis, which yields problems such as home bias, misallocation of capital, pro-cyclicality of capital flows, and vulnerability to sudden stops?

The home bias literature has shown that foreign investors often invest in only a small set of firms in an emerging market. As an example, while there are over 5000 listed firms in India, in 2011 there were only 703 firms where foreign investors owned above 5 per cent of the publicly traded (i.e. ‘floating’) market value. This raises questions about these chosen firms. What is the process of portfolio formation adopted by foreign investors? Do foreign investors possess a strong investment technology, through which their capital is channelled into good projects?

Numerous papers have been written in this literature, with often contradictory results. This paper takes the novel approach of separating the questions of asset allocation and security selection. This is a standard idea in the economics of fund management, going back to the 1960s. Portfolio returns can be decomposed into exposure to systematic asset pricing factors, such as size or book-to-market, as opposed to returns to security selection. Differences in asset allocation reflect the portfolio strategy of the investor, and there can be legitimate reasons for differences in exposures to asset pricing factors. In contrast, performance obtained through security selection unambiguously reflects investment technology.

We analyse the behaviour of foreign versus domestic institutional investors in India and find substantial differences in asset allocation. In some respects, foreign investors take on more risk, and should therefore obtain higher expected returns. In other respects, this operates in reverse; foreign investors take on reduced risk.

We then turn to the question of security selection. After controlling for differences in asset allocation, do foreign investors do well in choosing securities? Specifically, do firms chosen by foreign investors exhibit superior stock market returns? We look beyond the emphasis on returns in the finance perspective to also examine firm fundamentals. Do the firms chosen by foreign investors do better on growth in output, and growth of productivity?

All these outcomes of interest reflect a mix of a selection process (do foreign investors forecast well, and manage to identify firms that are going to do well?) and a treatment effect (does the decision by a foreign investor to buy shares in a company help the company do better?). We pursue the reduced form outcome, and make no attempt to disentangle selection from treatment effects.

We devise a quasi-experimental strategy for measuring the ability of foreign or domestic investors to do security selection, after controlling for differences in asset allocation. This involves identifying and addressing numerous threats to validity. Differences between firms in systematic asset pricing factors, such as size, B/P and β, are correlated with future outcomes. As an example, high β firms are likely to see high output growth in a business cycle expansion. In order to measure security selection, firms with high foreign institutional investment (but not domestic institutional investment) are matched against firms which got neither. Controls are identified which have similar size, B/P and β to the chosen firms. The comparison of outcomes identifies the security selection process, without being confounded by differences in asset allocation.

Our results may be summarised as follows. The firms chosen by foreign investors are those that have experienced high growth of capital (when compared with the control) prior to the observation date. They continue to obtain high growth of capital after the observation date. There is some evidence of superior output growth. However, the chosen firms have inferior productivity growth, and deliver inferior stock market returns when compared with the controls.

In contrast, the firms chosen by domestic institutional investors appear to deliver superior returns, and superior productivity growth, in the years after measurement date. This suggests that domestic institutions possess a valuable investment technology.

The methodology and the results of this paper have many implications. The literature on investment technology of foreign versus domestic investors, which has generally emphasised reduced form portfolio returns, has inconclusive results. Our results suggest that differences in the overall portfolio returns reflect a combination of differences in asset allocation and differences in security selection, which may explain how different researchers have obtained different results on the superiority of the investment technology of foreign investors. For foreign investors in India, these results suggest that the returns drag associated with poor capabilities in security selection could be avoided by achieving the desired asset allocation through index funds that express systematic asset pricing factors. The methodology of this paper can be easily extended to other countries, since the data requirements are easily met in all emerging markets.

The remainder of this paper is organised as follows. Section 2 sketches the questions and the measurement strategy. Section 3 describes the dataset used in the paper. Section 4 shows reduced form models explaining FII and DII ownership, which helps us obtain greater intuition into the phenomena at work. Section 5 examines the asset allocation choices of foreign and domestic institutional investors and finds substantial differences between the two. Section 6 measures the security selection process, after controlling for differences in asset allocation. Section 7 undertakes a series of modifications to the analysis in order to gauge the sensitivity of the results. Finally, Section 8 concludes.

2 Questions and methodology

The presence of home bias is a well established fact in the international finance literature: foreign investors hold relatively low weights in emerging markets. They tend to invest in large and liquid firms with international visibility and better corporate governance.

One strand of this literature has asked the question: Do foreign investors perform well? The presence of home bias, which suggests limitations in the information processing of foreign investors, may imply inferior investment performance by foreign investors. If the investment technology of foreign investors has difficulties, this could encourage investment in index funds that give exposure to emerging markets without engaging in security selection.

These questions are important to the policy debates about financial globalisation. If foreign investors suffer from asymmetric information and thus possess an inferior investment technology, their decisions could induce misallocation. Some of the pathologies identified by the international finance literature, such as the pro-cyclical behaviour of foreign investment or the phenomena of sudden stops and capital flow reversals, could be attributed to poor information processing by foreign investors.

This motivates a careful examination of the investment technology of foreign investors. The existing literature does not have a single unifying model and methodology; a series of papers have obtained diverse datasets, and each has fashioned a methodology suited to the dataset at hand. Dvořák (2005) utilises transaction data from the Jakarta Stock Exchange, and finds that clients of local brokerage firms do well in the short run, but clients of foreign brokerage firms do better in the long run. Choe, Kho, and Stulz (2005) find that foreign investors suffer higher transactions costs in Korea. Froot and Ramadorai (2008) harness a unique identification opportunity by juxtaposing closed-end country fund NAV returns and home country returns. They argue that institutional cross-border flows are based on sound information processing about country fundamentals. Albuquerque, H Bauer, and Schneider (2009) argue that the returns-chasing behaviour of US investors can be attributed to superior information, not inferior knowledge or trend-following.

The mainstream finance literature on these questions has focused on investment technology in the sense of returns forecasting. We broaden the analysis to also evaluate forecasts of firm fundamentals. A recent paper which has embarked on similar questions with the same dataset is Petkova (2012). As the home bias literature has demonstrated, foreign investors invest in only a small set of firms in an emerging markets. How well does this selectivity process work? Do the firms chosen by foreign investors do well in terms of growth of output and productivity?

While these questions are interesting and important, the analysis faces numerous threats to validity which need to be factored in while constructing a measurement strategy.

3 Data description

The dataset for our analysis is drawn from the CMIE Prowess database, from 2001 to 2011. This is a rich database where a wide array of information about large firms in India is observed. The industry structure of the dataset is shown in Table 1. As this table shows, the firms in our dataset are drawn from a diverse array of industries, and include many services firms also. Table 2 shows summary statistics about the firms in the dataset.

Table 1

Industry Composition

This table shows the number of firms in each major industry group, in each year, of the dataset under examination. In addition to manufacturing firms, we also observe many services firms. As an example, there were 155 information technology firms in 2001, which went up to 206 in 2011.

Table 1

Industry Composition

This table shows the number of firms in each major industry group, in each year, of the dataset under examination. In addition to manufacturing firms, we also observe many services firms. As an example, there were 155 information technology firms in 2001, which went up to 206 in 2011.

	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011
Chemicals	482	509	551	516	522	506	578	574	578	570	571
Diversified	33	34	34	31	34	38	40	42	42	42	42
Electricity	11	12	12	13	15	17	17	21	22	23	24
Food	193	211	223	210	210	208	247	252	251	246	252
Machinery	254	273	264	268	267	265	287	288	284	279	283
Metals	179	199	203	202	204	204	244	249	249	241	243
Mining	18	21	22	20	22	23	26	26	28	29	32
MiscManuf	101	108	106	104	102	111	136	137	138	141	139
NonMetalMin	114	127	132	125	124	121	135	135	138	137	142
Serv.Construction	109	116	112	110	116	124	141	160	166	175	191
Serv.IT	155	151	167	153	154	154	182	195	195	200	206
Serv.Other	460	480	479	444	455	475	544	543	529	537	578
Textiles	248	274	285	272	266	264	307	305	298	292	299
TransportEq	90	95	92	90	98	100	113	113	117	117	115
Sum	2447	2610	2682	2558	2589	2610	2997	3040	3035	3029	3117

View Table

Table 1

Industry Composition

This table shows the number of firms in each major industry group, in each year, of the dataset under examination. In addition to manufacturing firms, we also observe many services firms. As an example, there were 155 information technology firms in 2001, which went up to 206 in 2011.

	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011
Chemicals	482	509	551	516	522	506	578	574	578	570	571
Diversified	33	34	34	31	34	38	40	42	42	42	42
Electricity	11	12	12	13	15	17	17	21	22	23	24
Food	193	211	223	210	210	208	247	252	251	246	252
Machinery	254	273	264	268	267	265	287	288	284	279	283
Metals	179	199	203	202	204	204	244	249	249	241	243
Mining	18	21	22	20	22	23	26	26	28	29	32
MiscManuf	101	108	106	104	102	111	136	137	138	141	139
NonMetalMin	114	127	132	125	124	121	135	135	138	137	142
Serv.Construction	109	116	112	110	116	124	141	160	166	175	191
Serv.IT	155	151	167	153	154	154	182	195	195	200	206
Serv.Other	460	480	479	444	455	475	544	543	529	537	578
Textiles	248	274	285	272	266	264	307	305	298	292	299
TransportEq	90	95	92	90	98	100	113	113	117	117	115
Sum	2447	2610	2682	2558	2589	2610	2997	3040	3035	3029	3117

Table 2

Summary Statistics

Some features of the dataset are shown in the table.

Table 2

Summary Statistics

Some features of the dataset are shown in the table.

Variable	Units	Mean	SD	Min	25th	Median	75th	Max
Sales	Rs. million	7153.61	63329.24	1.00	141.65	645.50	2655.40	3574219.80
Gross Fixed Assets	Rs. million	4255.86	36627.87	1.00	96.90	371.30	1486.70	2212519.70
Total Assets	Rs.million	7943.24	55635.54	1.00	180.50	695.00	2892.00	2849003.50
Employees	Number	1227.04	5342.23	1.00	35.57	183.07	790.70	159999.57
Wage bill	Rs. million	337.15	2669.62	0.10	6.50	30.20	130.10	134173.50
FII ownership	Per cent	4.05	10.85	0.00	0.00	0.00	0.36	89.29
DII ownership	Per cent	9.60	15.66	0.00	0.00	1.22	13.48	100.00
Promoter ownership	Per cent	49.01	20.01	0.00	35.64	50.03	63.27	99.83
Adjusted Closing Price	Rs.	68.85	378.85	0.01	4.00	12.35	45.35	49088.80
TFP (Levinsohn Petrin)		21.25	8.25	1.00	18.70	20.68	22.49	186.55

View Table

Table 2

Summary Statistics

Some features of the dataset are shown in the table.

Variable	Units	Mean	SD	Min	25th	Median	75th	Max
Sales	Rs. million	7153.61	63329.24	1.00	141.65	645.50	2655.40	3574219.80
Gross Fixed Assets	Rs. million	4255.86	36627.87	1.00	96.90	371.30	1486.70	2212519.70
Total Assets	Rs.million	7943.24	55635.54	1.00	180.50	695.00	2892.00	2849003.50
Employees	Number	1227.04	5342.23	1.00	35.57	183.07	790.70	159999.57
Wage bill	Rs. million	337.15	2669.62	0.10	6.50	30.20	130.10	134173.50
FII ownership	Per cent	4.05	10.85	0.00	0.00	0.00	0.36	89.29
DII ownership	Per cent	9.60	15.66	0.00	0.00	1.22	13.48	100.00
Promoter ownership	Per cent	49.01	20.01	0.00	35.64	50.03	63.27	99.83
Adjusted Closing Price	Rs.	68.85	378.85	0.01	4.00	12.35	45.35	49088.80
TFP (Levinsohn Petrin)		21.25	8.25	1.00	18.70	20.68	22.49	186.55

Not all firms in the CMIE database disclose the number of employees. In this paper, we have used data within a given industry within a year, to compute the average wage using firms where the number of employees was observed. This was used to impute the number of employees for observations where the overall wages was observed but the number of firms was not observed.

One simple measure of productivity is obtained by differentiating the Cobb-Douglas production function: $\dot{w}=\dot{y}-{\mathit{\alpha }}_l\dot{l}-\left(1-{\mathit{\alpha }}_l\right)\dot{k}$ where ẇ is the productivity growth, y is log of output, k is log of capital, l is log of labour and α_l is the share of labour i.e. the ratio of wages to sales.

In the estimation of firm productivity, Levinsohn and Petrin (2003) argued that there is an endogeneity problem owing to the correlation between unobservable productivity shocks and the input levels. They propose using intermediate inputs (raw material expenditure) as a proxy for the purpose of estimation. We implement their procedure using their Stata code. This procedure cannot be implemented for services firms which do not buy raw materials. Hence, when we analyse productivity using the Levinsohn-Petrin measure, we lose data for services firms.

Finally, we turn to describing categories of institutional investors observed in this dataset. Indian capital controls only permit registered ‘foreign institutional investors’ (FIIs) to invest in the equity market. Once registration is done, the investment process is fairly unconstrained. Three kinds of domestic institutional investors (DIIs) are present – banks, mutual funds and insurance companies. We focus on the ownership by FIIs and by DIIs.

Most firms in India have a dominant manager/shareholder, which is typically a family, which retains strategic control of the firm for very long time horizons. In the Indian parlance, this shareholder is termed ‘the promoter’. As Table 2 shows, the median firm has promoter ownership of 50.03 per cent, i.e. full control.

Institutional investors can only choose to hold shares which are not held by the dominant shareholder. Hence, we rescale the observed shareholding by FIIs and DIIs by the total outsider shareholding. As an example, if the promoter owns 60%, and if FIIs own 20%, then we rescale this to 50%. That is, for this firm, FIIs own half the shares traded in the public market.

The median firm in the dataset has no foreign institutional investor (FII) shareholding. At the 75th percentile, FII ownership is at 0.36%. Most firms appear to be shunned by foreign investors. The median firm has just 1.22% of ownership by DIIs. At the 75th percentile, we see DII shareholding of 13.48 per cent. While this is bigger than the 75th percentile ownership of FIIs, it remains a small number. For three quarters of the firms, DIIs own below 13.48 per cent of the shares relinquished by insiders.

This raises questions about institutional ownership. What firms are chosen by FIIs and DIIs? How effective is the process of selection employed by these investors? Are institutional investors able to identify the firms with the best prospects, who would achieve high growth and stock market returns in the future? We explore these questions in the remainder of this paper.

4 A preliminary exploration

The first exploration that we must embark on is to examine the process of security selection of FIIs and DIIs. If, hypothetically, we find that FIIs and DIIs behave similarly, then there is no need for further exploration. As an example, Dahlquist and Robertsson (2001) find that in Sweden, while there are differences between foreign investment and domestic investment, these derive primarily through the fact that most foreign investment is institutional. When foreign institutional investors are compared against Swedish institutional investors, their preferences for firms are largely alike.

Table 3 shows that DIIs have a much bigger shareholding in top quintile companies by asset tangibility, while no strong pattern is visible with FIIs. Similarly, Table 4 shows that DIIs appear to have a lower shareholding for firms with bottom quintile values for insider shareholding.

Table 3

Institutional ownership by asset tangibility

This table breaks down the dataset into quintiles by asset tangibility, defined as the fraction of total assets which are tangible assets. In the bottom quintile, this has a median value of 14.05%, while in the top quintile, this has a median value of 96.02%. In each quintile, we report the median value of FII and DII ownership. The median DII ownership is much higher (20.33%) in the top quintile.

Table 3

Institutional ownership by asset tangibility

This table breaks down the dataset into quintiles by asset tangibility, defined as the fraction of total assets which are tangible assets. In the bottom quintile, this has a median value of 14.05%, while in the top quintile, this has a median value of 96.02%. In each quintile, we report the median value of FII and DII ownership. The median DII ownership is much higher (20.33%) in the top quintile.

	Q1	Q2	Q3	Q4	Q5
Tangibility	14.05	34.74	51.10	69.32	96.02
FII	0.00	0.00	0.00	0.00	0.89
DII	0.76	0.27	1.39	2.92	20.33

View Table

Table 3

Institutional ownership by asset tangibility

This table breaks down the dataset into quintiles by asset tangibility, defined as the fraction of total assets which are tangible assets. In the bottom quintile, this has a median value of 14.05%, while in the top quintile, this has a median value of 96.02%. In each quintile, we report the median value of FII and DII ownership. The median DII ownership is much higher (20.33%) in the top quintile.

	Q1	Q2	Q3	Q4	Q5
Tangibility	14.05	34.74	51.10	69.32	96.02
FII	0.00	0.00	0.00	0.00	0.89
DII	0.76	0.27	1.39	2.92	20.33

Table 4

Institutional ownership by insider holding

This table breaks down the dataset into quintiles by insider shareholding. Bottom quintile companies, by insider shareholding, have a median insider shareholding of 25.57%. Top quintile companies, by insider shareholding, have a median insider shareholding of 73.46%. In all quintiles, the median value of FII ownership is 0. In the case of DIIs, the median value is lower for Q1 when compared with the other quintiles, i.e. DIIs appear to shun companies where a dominant shareholder has a relatively small shareholding.

Table 4

Institutional ownership by insider holding

This table breaks down the dataset into quintiles by insider shareholding. Bottom quintile companies, by insider shareholding, have a median insider shareholding of 25.57%. Top quintile companies, by insider shareholding, have a median insider shareholding of 73.46%. In all quintiles, the median value of FII ownership is 0. In the case of DIIs, the median value is lower for Q1 when compared with the other quintiles, i.e. DIIs appear to shun companies where a dominant shareholder has a relatively small shareholding.

	Q1	Q2	Q3	Q4	Q5
Insider holding	25.57	40.99	50.96	59.99	73.46
FII	0.01	0.02	0.02	0.00	0.00
DII	0.74	3.53	4.24	3.55	2.60

View Table

Table 4

Institutional ownership by insider holding

This table breaks down the dataset into quintiles by insider shareholding. Bottom quintile companies, by insider shareholding, have a median insider shareholding of 25.57%. Top quintile companies, by insider shareholding, have a median insider shareholding of 73.46%. In all quintiles, the median value of FII ownership is 0. In the case of DIIs, the median value is lower for Q1 when compared with the other quintiles, i.e. DIIs appear to shun companies where a dominant shareholder has a relatively small shareholding.

	Q1	Q2	Q3	Q4	Q5
Insider holding	25.57	40.99	50.96	59.99	73.46
FII	0.01	0.02	0.02	0.00	0.00
DII	0.74	3.53	4.24	3.55	2.60

In order to explore these relationships, we wish to estimate linear models explaining FII and DII ownership in terms of firm characteristics. Summary statistics about firm characteristics of interest are shown in Table 5. Many firms have zero values for either or both of these. Hence, we resort to Tobit models. Clustered standard errors are reported to reflect the fact that a given firm can often be observed in many years.

Table 5

Firm characteristics that may influence FII and DII ownership

This table shows summary statistics about firm characteristics that may influence FII and DII ownership. Turnover ratio is the latest 12 month turnover expressed as a ratio of market capitalisation. The largest value, of 3.45, denotes turnover which is 3.45 times the market capitalisation. Yield is the dividend yield expressed in per cent. Total risk is the standard deviation of daily returns. Age is measured in years. Asset tangibility is the tangible assets expressed as per cent of total assets. R&D intensity is the expense on R&D expressed as per cent of sales.

Table 5

Firm characteristics that may influence FII and DII ownership

This table shows summary statistics about firm characteristics that may influence FII and DII ownership. Turnover ratio is the latest 12 month turnover expressed as a ratio of market capitalisation. The largest value, of 3.45, denotes turnover which is 3.45 times the market capitalisation. Yield is the dividend yield expressed in per cent. Total risk is the standard deviation of daily returns. Age is measured in years. Asset tangibility is the tangible assets expressed as per cent of total assets. R&D intensity is the expense on R&D expressed as per cent of sales.

Variable	Mean	SD	Min	25th	Median	75th	Max	IQR	Observations
Yield	1.92	3.29	0.00	0.00	0.00	2.68	16.73	2.68	25402.00
Global beta	0.64	0.66	−6.47	0.29	0.63	0.95	7.61	0.67	15250.00
Total Risk	0.85	0.47	0.26	0.56	0.72	0.96	2.86	0.40	20251.00
Export-Sales ratio	15.97	26.50	0.00	0.00	1.68	19.38	100.00	19.38	28155.00
Age	25.95	18.03	1.00	15.00	20.00	30.00	148.00	15.00	30773.00
Tangibility	63.13	43.77	1.27	32.38	56.69	84.51	244.94	52.12	29101.00
R and D intensity	0.23	0.89	0.00	0.00	0.00	0.00	6.77	0.00	28243.00

View Table

Table 5

Firm characteristics that may influence FII and DII ownership

This table shows summary statistics about firm characteristics that may influence FII and DII ownership. Turnover ratio is the latest 12 month turnover expressed as a ratio of market capitalisation. The largest value, of 3.45, denotes turnover which is 3.45 times the market capitalisation. Yield is the dividend yield expressed in per cent. Total risk is the standard deviation of daily returns. Age is measured in years. Asset tangibility is the tangible assets expressed as per cent of total assets. R&D intensity is the expense on R&D expressed as per cent of sales.

Variable	Mean	SD	Min	25th	Median	75th	Max	IQR	Observations
Yield	1.92	3.29	0.00	0.00	0.00	2.68	16.73	2.68	25402.00
Global beta	0.64	0.66	−6.47	0.29	0.63	0.95	7.61	0.67	15250.00
Total Risk	0.85	0.47	0.26	0.56	0.72	0.96	2.86	0.40	20251.00
Export-Sales ratio	15.97	26.50	0.00	0.00	1.68	19.38	100.00	19.38	28155.00
Age	25.95	18.03	1.00	15.00	20.00	30.00	148.00	15.00	30773.00
Tangibility	63.13	43.77	1.27	32.38	56.69	84.51	244.94	52.12	29101.00
R and D intensity	0.23	0.89	0.00	0.00	0.00	0.00	6.77	0.00	28243.00

The results of this estimation are shown in Table 6. In the case of FIIs, we find strong results where FIIs favour firms with reduced shareholding by insiders, bigger size, a high domestic beta, a high beta against a global index, low total risk and lower age. They avoid the dummy variable for public sector corporations. They favour reduced asset tangibility. Firms are broken into three groups by R&D expenses – None, low and high – and three dummy variables are constructed. Of these, ‘None’ is the ommitted category. FIIs appear to weakly favour firms with high R&D expenses.

Table 6

Tobit models that explain FII and DII ownership

We wish to explore the relationships between FII and DII ownership, and firm characteristics. Many firms have zero values for FII or DII investment (or both). Hence, we estimate Tobit models. Correlations within firm are addressed by clustered standard errors. Macroeconomic effects are controlled by having year fixed effects. The table shows estimation results for a tobit model explaining FII ownership and another tobit model explaining DII ownership. Both models use the identical set of explanatory variables.

As an example, the coefficient of insider shareholding is -0.13 (with a t statistic of -7.17) for the FII tobit, while it is -0.03 (with a t statistic of -1.77) for the DII tobit.

Table 6

Tobit models that explain FII and DII ownership

We wish to explore the relationships between FII and DII ownership, and firm characteristics. Many firms have zero values for FII or DII investment (or both). Hence, we estimate Tobit models. Correlations within firm are addressed by clustered standard errors. Macroeconomic effects are controlled by having year fixed effects. The table shows estimation results for a tobit model explaining FII ownership and another tobit model explaining DII ownership. Both models use the identical set of explanatory variables.

As an example, the coefficient of insider shareholding is -0.13 (with a t statistic of -7.17) for the FII tobit, while it is -0.03 (with a t statistic of -1.77) for the DII tobit.

	FII	t	DII	t
Insider holding	−0.13	−7.16	−0.02	−1.74
Log mktcap	7.14	30.01	3.88	22.05
Turnover ratio	0.39	1.14	−1.36	−5.48
Yield	−0.29	−3.25	−0.09	−1.36
Domestic beta	3.39	4.61	−0.49	−0.97
Global beta	0.74	1.77	−0.17	−0.51
Total Risk	−4.92	−2.80	−0.43	−0.35
Export to sales	0.01	1.11	−0.01	−0.81
Age	−0.11	−5.72	0.16	9.39
Is public sector	−6.00	−2.76	10.32	4.52
Tangibility	−0.03	−2.94	0.08	9.06
Low R and D	−0.33	−0.59	1.74	3.43
High R and D	0.99	1.64	−1.55	−2.66

View Table

Table 6

Tobit models that explain FII and DII ownership

We wish to explore the relationships between FII and DII ownership, and firm characteristics. Many firms have zero values for FII or DII investment (or both). Hence, we estimate Tobit models. Correlations within firm are addressed by clustered standard errors. Macroeconomic effects are controlled by having year fixed effects. The table shows estimation results for a tobit model explaining FII ownership and another tobit model explaining DII ownership. Both models use the identical set of explanatory variables.

As an example, the coefficient of insider shareholding is -0.13 (with a t statistic of -7.17) for the FII tobit, while it is -0.03 (with a t statistic of -1.77) for the DII tobit.

	FII	t	DII	t
Insider holding	−0.13	−7.16	−0.02	−1.74
Log mktcap	7.14	30.01	3.88	22.05
Turnover ratio	0.39	1.14	−1.36	−5.48
Yield	−0.29	−3.25	−0.09	−1.36
Domestic beta	3.39	4.61	−0.49	−0.97
Global beta	0.74	1.77	−0.17	−0.51
Total Risk	−4.92	−2.80	−0.43	−0.35
Export to sales	0.01	1.11	−0.01	−0.81
Age	−0.11	−5.72	0.16	9.39
Is public sector	−6.00	−2.76	10.32	4.52
Tangibility	−0.03	−2.94	0.08	9.06
Low R and D	−0.33	−0.59	1.74	3.43
High R and D	0.99	1.64	−1.55	−2.66

The results for DIIs are strikingly different. While FIIs avoid firms with a high inside shareholding, this does not influence DIIs. The coefficient for size is much weaker: DIIs invest in smaller firms than FIIs. DIIs strongly avoid illiquid stocks while FIIs do not care about stock market liquidity. While FIIs favour domestic β exposure, DIIs are not influenced by it, or by the global β. While FIIs avoid total risk, DIIs are not concerned about it.

Like FIIs, DIIs are not influenced by the exports/sales ratio. They strongly favour older companies, in contrast to FIIs who favour young companies. DIIs own much more public sector companies, while FIIs systematically avoid them. DIIs strongly favour firms with more tangible assets, while DIIs favour firms with reduced tangible assets. Finally, DIIs invest more in low R&D companies and avoid firms that do more R&D.

If we believe that dynamic companies are young, private, with low tangible assets, and high R&D, then it appears that FIIs systematically favour these firms while DIIs shun them. While it may be obvious that young or private or high R&D companies are good, they may not achieve high stock market returns or growth in the future. The information processing capabilities of an investor must be evaluated by examining the performance of firms in the period after investment date. The investment technology of an institutional investor can be evaluated in two ways: by comparing stock market returns in the future, and also by comparing the economic performance of firms in the future. However, in undertaking these comparisons, we have to be conscious of differences in asset allocation and control for these.

5 The asset allocation of FIIs and DIIs

The empirical asset pricing literature has emphasised factors: size (log of market capitalisation), book-to-price and β (Fama and French, 1993). The expected returns of a portfolio tend to be higher when it tilts towards high β, low size and high book-to-price firms. Summary statistics for these firm characteristics are shown in Table 7.²

Table 7

Summary statistics about asset pricing characteristics

This table shows summary statistics about firm characteristics of interest. As an example, log market capitalisation ranged from -1.14 to 15.07 with a median value of 5.55.

Table 7

Summary statistics about asset pricing characteristics

This table shows summary statistics about firm characteristics of interest. As an example, log market capitalisation ranged from -1.14 to 15.07 with a median value of 5.55.

Variable	Mean	SD	Min	25th	Median	75th	Max	IQR	Observations
Domestic Beta	0.82	0.50	−2.50	0.54	0.80	1.07	5.31	0.53	15882.00
Log Book-to-Price	0.11	1.19	−7.06	−0.66	0.14	0.89	4.61	1.55	22989.00
Size	5.82	2.41	−1.14	4.00	5.55	7.35	15.07	3.35	25402.00
Log Momentum	1.59	0.09	−0.12	1.56	1.61	1.62	2.43	0.06	28523.00
Turnover Ratio	0.22	0.56	0.00	0.01	0.04	0.16	3.97	0.15	24916.00

View Table

Table 7

Summary statistics about asset pricing characteristics

This table shows summary statistics about firm characteristics of interest. As an example, log market capitalisation ranged from -1.14 to 15.07 with a median value of 5.55.

Variable	Mean	SD	Min	25th	Median	75th	Max	IQR	Observations
Domestic Beta	0.82	0.50	−2.50	0.54	0.80	1.07	5.31	0.53	15882.00
Log Book-to-Price	0.11	1.19	−7.06	−0.66	0.14	0.89	4.61	1.55	22989.00
Size	5.82	2.41	−1.14	4.00	5.55	7.35	15.07	3.35	25402.00
Log Momentum	1.59	0.09	−0.12	1.56	1.61	1.62	2.43	0.06	28523.00
Turnover Ratio	0.22	0.56	0.00	0.01	0.04	0.16	3.97	0.15	24916.00

Table 8 repeats the Tobit analysis, where the explanatory variables are quartile dummies for the size, B/P, and β. The results differ strongly, which suggests that FIIs and DIIs differ strongly in their exposure to empirical asset pricing factors. These differences need to be controlled for when examining the future performance of firms.

Table 8

Tobit model based on empirical asset pricing characteristics

This table shows Tobit models explaining FII and DII ownership based on empirical asset pricing factors. For each of the three factors – size, B/P and β – we construct four quartile dummies. The results show that FIIs and DIIs differ strongly in their asset pricing exposures.

Table 8

Tobit model based on empirical asset pricing characteristics

This table shows Tobit models explaining FII and DII ownership based on empirical asset pricing factors. For each of the three factors – size, B/P and β – we construct four quartile dummies. The results show that FIIs and DIIs differ strongly in their asset pricing exposures.

	FII	t	DII	t
Intercept	−8.75	−16.81	5.41	19.81
Small size	30.49	24.35	20.10	25.95
Med size	6.92	9.44	4.71	9.32
Large size	1.38	2.78	1.50	3.99
Low Book-to-Price	1.56	2.47	5.56	9.14
Med Book-to-Price	4.99	10.23	2.74	6.52
High Book-to-Price	−0.42	−1.14	−0.03	−0.10
Low beta	4.63	7.51	−0.15	−0.33
Med beta	−0.85	−1.80	−0.10	−0.27
High beta	−0.01	−0.02	0.47	1.51

View Table

Table 8

Tobit model based on empirical asset pricing characteristics

This table shows Tobit models explaining FII and DII ownership based on empirical asset pricing factors. For each of the three factors – size, B/P and β – we construct four quartile dummies. The results show that FIIs and DIIs differ strongly in their asset pricing exposures.

	FII	t	DII	t
Intercept	−8.75	−16.81	5.41	19.81
Small size	30.49	24.35	20.10	25.95
Med size	6.92	9.44	4.71	9.32
Large size	1.38	2.78	1.50	3.99
Low Book-to-Price	1.56	2.47	5.56	9.14
Med Book-to-Price	4.99	10.23	2.74	6.52
High Book-to-Price	−0.42	−1.14	−0.03	−0.10
Low beta	4.63	7.51	−0.15	−0.33
Med beta	−0.85	−1.80	−0.10	−0.27
High beta	−0.01	−0.02	0.47	1.51

Figure 1 juxtaposes the exposure of FIIs and DIIs to the three asset pricing factors and Table 9 shows their median exposure from 2001 to 2011.

View Full Size

Exposure to asset pricing factors

We examine the FII and DII exposure to systematic asset pricing factors in this figure. The exposure is calculated as the weighted sum of the asset pricing factor, where the weight is equal to the funds allocated to a security as a percentage of the portfolio. For example, in a two-firm world, if the FII ownership is w₁ and w₂, market capitalisation is m₁ and m₂, and the firm beta is b₁ and b₂. Then weight of each security is x₁ = w₁m₁/(w₁m₁ + w₂m₂) and the FII portfolio beta is equal to x₁b₁ + x₂b₂.

Citation: IMF Working Papers 2013, 090; 10.5089/9781484335567.001.A001

• Download Figure
• Download figure as PowerPoint slide

Table 9

Median exposure to asset pricing factors

Table 9

Median exposure to asset pricing factors

	Beta	Size	Log Book-to-price
FII	0.93	11.05	−1.37
DII	0.73	9.17	−0.80
Difference	0.20	1.88	−0.58

View Table

Table 9

Median exposure to asset pricing factors

	Beta	Size	Log Book-to-price
FII	0.93	11.05	−1.37
DII	0.73	9.17	−0.80
Difference	0.20	1.88	−0.58

This shows substantial differences across the asset allocation of these two investors. In most years, FIIs carry a higher beta exposure than the DIIs, with a 20 points higher median exposure over our sample period. A higher beta exposure suggests higher returns. FIIs as compared to DIIs also carry a higher size exposure from 2001 to 2008, and a lower size exposure from 2009 to 2011. But the median size exposure over the sample period is higher by 1.88 points, suggeting lower returns for the FIIs. In terms of book-to-price, FIIs are exposed to low book-to-price firms in all periods with a median difference of 0.58 points. Low book to price exposure suggests lower returns.

There is a systematic difference in the asset allocation by FIIs and DIIs and this confirms the earlier argument, that these differences need to be controlled for when examining the future performance of stock returns or firm fundamentals; these may diverge owing to asset allocation and not security selection.

6 The security selection of FIIs and DIIs

We would like to judge the investment technology of FIIs and DIIs by evaluating the performance of firms in the future, after the date on which the shareholding pattern is observed. At the same time, we would like to control for differences in exposures to empirical asset pricing factors. This would help us assess the security selection by FIIs and DIIs, without being confounded by differences in asset allocation, i.e. systematic asset pricing factors.

We assign firm years to the following groups based on the median ownership of firms by FIIs and DIIs. We define a ‘High FII’ group as one where FII ownership was above 5%, and DII ownership was below 6%. These are the firms favoured by FIIs but disfavoured by DIIs. Similarly, we define a ‘High DII’ group where DII ownership was above 6%, but FII ownership was below 5%. Finally, a control pool is constructed of firms where neither FII nor DII ownership was above their median values (i.e. 5% for FII ownership and 6% for DII ownership). We would like to compare the future performance of a High FII company against a similar company from ‘None’, and the future performance of a High DII company against a similar company from ‘None’.

Table 10 shows the number of firms falling into the four categories (‘Both’, ‘High DII’, ‘High FII’ and ‘None’) in each year. In 2011, there were 483 ‘High DII’ firms and 274 ‘High FII’ firms. There was a large number of firms in ‘None’, the control pool.

Table 10

Number of firms in each category

The table shows the number of firms in each year, which fall into the four categories ‘Both’ (investment by both FIIs and DIIs that is above-median), ‘High DII’ (above-median investment by DIIs but below-median investment by FIIs), ‘High FII’ (above-median investment by FIIs but below-median investment by DIIs) and ‘None’ (below-median investment by both FIIs and DIIs).

Table 10

Number of firms in each category

The table shows the number of firms in each year, which fall into the four categories ‘Both’ (investment by both FIIs and DIIs that is above-median), ‘High DII’ (above-median investment by DIIs but below-median investment by FIIs), ‘High FII’ (above-median investment by FIIs but below-median investment by DIIs) and ‘None’ (below-median investment by both FIIs and DIIs).

	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011
Both	181	141	134	185	241	375	442	495	416	421	428
High DII	927	980	962	807	708	598	635	572	593	539	483
High FII	35	41	38	63	97	141	207	246	262	236	274
None	1319	1466	1560	1508	1547	1496	1719	1732	1769	1838	1937
Sum	2462	2628	2694	2563	2593	2610	3003	3045	3040	3034	3122

View Table

Table 10

Number of firms in each category

The table shows the number of firms in each year, which fall into the four categories ‘Both’ (investment by both FIIs and DIIs that is above-median), ‘High DII’ (above-median investment by DIIs but below-median investment by FIIs), ‘High FII’ (above-median investment by FIIs but below-median investment by DIIs) and ‘None’ (below-median investment by both FIIs and DIIs).

	2001	2002	2003	2004	2005	2006	2007	2008	2009	2010	2011
Both	181	141	134	185	241	375	442	495	416	421	428
High DII	927	980	962	807	708	598	635	572	593	539	483
High FII	35	41	38	63	97	141	207	246	262	236	274
None	1319	1466	1560	1508	1547	1496	1719	1732	1769	1838	1937
Sum	2462	2628	2694	2563	2593	2610	3003	3045	3040	3034	3122

Table 11 shows transition probabilities on a one year horizon across these four categories. We observe that DII investment is sticky, but FIIs change portfolio frequently. There is a strong possibility of dropping back to ‘None’ in year t + 1 after being in either ‘High DII’ or ‘High FII’ category at time t. Once a firm is in ‘High FII’ category, there is an 18.46% chance that it will drop into ‘None’ in the next year, but there is a 12.33% chance that it will go up to ‘Both’ in the next year by gaining high DII investment also.

Table 11

Transition probabilities across the four groups of firms

Each row of this table shows probabilities for where a firm would be in year t + 1 given that it is in a certain category in year t. As an example, a firm which is classified as ‘None’ at time t would stay in that state in year t + 1 with a probability of 94.98%. There is a 2.26% per cent chance that it would jump up to ‘High FII’ and a 2.13% chance that it would jump up to ‘High DII’. Once it goes into ‘High FII’ in time t, there is a 18.46% chance of it falling back to ‘None’.

Table 11

Transition probabilities across the four groups of firms

Each row of this table shows probabilities for where a firm would be in year t + 1 given that it is in a certain category in year t. As an example, a firm which is classified as ‘None’ at time t would stay in that state in year t + 1 with a probability of 94.98%. There is a 2.26% per cent chance that it would jump up to ‘High FII’ and a 2.13% chance that it would jump up to ‘High DII’. Once it goes into ‘High FII’ in time t, there is a 18.46% chance of it falling back to ‘None’.

	Both	High DII	High FII	None
Both	80.50	10.68	6.81	2.00
High DII	5.86	82.54	0.91	10.69
High FII	12.33	1.72	67.49	18.46
None	0.63	2.13	2.26	94.98

View Table

Table 11

Transition probabilities across the four groups of firms

Each row of this table shows probabilities for where a firm would be in year t + 1 given that it is in a certain category in year t. As an example, a firm which is classified as ‘None’ at time t would stay in that state in year t + 1 with a probability of 94.98%. There is a 2.26% per cent chance that it would jump up to ‘High FII’ and a 2.13% chance that it would jump up to ‘High DII’. Once it goes into ‘High FII’ in time t, there is a 18.46% chance of it falling back to ‘None’.

	Both	High DII	High FII	None
Both	80.50	10.68	6.81	2.00
High DII	5.86	82.54	0.91	10.69
High FII	12.33	1.72	67.49	18.46
None	0.63	2.13	2.26	94.98

While there are significant differences between the exposures of FII and DII and seen in figure 1, it will be possible to find firms in ‘None’ which are similar in terms of the asset pricing factors to those seen in ‘High FII’ and ‘High DII’ groups, so as to undertake comparisons.

At the simplest, an OLS model explaining an outcome of interest y_it (such as stock market returns or sales growth or productivity) could control for size, B/P and β:

y_i,t+j − y_i,t−k = β₀ + β₁size_i,t + β₂book/price_i,t + β₃beta_i,t + γ′D_i,t + e_it

In this regression, we are interested in the coefficients γ about membership in the group ‘High FII’ or ‘High DII’ in year t − 1. Differences between firms in size, B/P and β. would be controlled for. We utilise information about the investment characteristics at time t in order to make statements about the growth of an outcome variable y from year t − k to year t + j.

This traditional regression strategy suffers from certain weaknesses (Stuart, 2010). The impact of size, B/P and β upon y might not be linear. When the design matrix involves some firms in ‘None’ and some firms in ‘High FII’, and their characteristics differ, the OLS regression relies upon linear extrapolation to overcome these differences. This extrapolation is fraught with difficulties. A design matrix constructed with all observations may be a poor path to sound estimates of γ.

In order to address these problems, we propose a matching-based strategy. For each firm in the ‘High FII’ or ‘High DII’ categories, we use Mahalanobis distance matching in order to find a matched partner from the ‘None’ category, aiming to match on size, B/P and β.³ For each of these categories, this would yield a balanced design. A caliper is used, to delete observations where the match quality is poor. Matched pairs from all years are pooled to construct the dataset where OLS estimation is done:

y_i,t+j − y_i,t−k = β₀ + γD_i,t + e_i,t

We utilise information about the investment characteristics at time t in order to make statements about the growth of an outcome variable y from year t − k to year t + j.

The design here is a series of matched pairs. For the firm with ‘High FII’, we have D = 1, while its matched partner (which has neither high FII nor high DII investment), D = 0. The coefficient γ thus reports on the extent to which selection by FIIs at time t impacts upon the outcome y.

It is also interesting to look backwards into time. The left hand side variable y_i,t can pertain to conditions that prevailed before date t. In this case, the results of the regression give us insights into the process of selectivity by FIIs or DIIs. At the same time, results would be different, when compared with those obtained through the tobit regressions above, since the design matrix here is a more carefully constructed one.

Table 12 shows the results of this matching process for firms with high FII investment. As an example, in 2002, there were 41 firms with high FII investment but low DII investment, and 1466 firms with low FII investment and low DII investment. However, Mahalanobis distance matching based on size, B/P and β yielded only 14 matches. Overall, we see that a fairly large dataset of matched pairs is assembled using this process.

Table 12

Number of matched pairs for high FII

Table 12

Number of matched pairs for high FII

	Low FII, Low DII	High FII, Low DII	Matched pairs
2002	1466	41	14
2003	1560	38	10
2004	1508	63	34
2005	1547	97	63
2006	1496	141	59
2007	1719	207	91
2008	1732	246	127
2009	1769	262	182
2010	1838	236	179
2011	1937	274	199

View Table

Table 12

Number of matched pairs for high FII

	Low FII, Low DII	High FII, Low DII	Matched pairs
2002	1466	41	14
2003	1560	38	10
2004	1508	63	34
2005	1547	97	63
2006	1496	141	59
2007	1719	207	91
2008	1732	246	127
2009	1769	262	182
2010	1838	236	179
2011	1937	274	199

The same strategy, applied to high DII investment firms (with low FII investment) yields matched pairs as shown in Table 13. Here, a much larger number of matched pairs is obtained.

Table 13

Number of matched pairs for high DII

Table 13

Number of matched pairs for high DII

	Low FII, Low DII	High DII, Low FII	Matched pairs
2002	1466	980	407
2003	1560	962	322
2004	1508	807	236
2005	1547	709	208
2006	1496	598	204
2007	1719	635	220
2008	1732	572	219
2009	1769	593	286
2010	1838	539	266
2011	1937	483	244

View Table

Table 13

Number of matched pairs for high DII

	Low FII, Low DII	High DII, Low FII	Matched pairs
2002	1466	980	407
2003	1560	962	322
2004	1508	807	236
2005	1547	709	208
2006	1496	598	204
2007	1719	635	220
2008	1732	572	219
2009	1769	593	286
2010	1838	539	266
2011	1937	483	244

6.1 Match balance

The first question that has to be addressed is about the extent to which this quasi-experimental strategy achieves match balance.

Table 14 shows standardised differences of size, B/P and β for firms with high FII investment (but low DII investment). The matching process has worked well; the standardised differences have dropped to near zero. This is reinforced by Kolmogorov-Smirnoff tests shown in Table 15. The null of equality of distributions is always rejected in the raw data and is never rejected after matching.

Table 14

Standardised Difference for FII

Table 14

Standardised Difference for FII

	Before Matching	After Matching
Size	1.46	0.05
Book-to-Price	−0.75	−0.02
Beta	0.30	0.05

View Table

Table 14

Standardised Difference for FII

	Before Matching	After Matching
Size	1.46	0.05
Book-to-Price	−0.75	−0.02
Beta	0.30	0.05

Table 15

Kolmogorov Smirnov Test for FII

Table 15

Kolmogorov Smirnov Test for FII

	Before Matching	After Matching
Size	0.5716	0.048
	(0)	(0.2194)
Book-to-Price	0.3061	0.0303
	(0)	(0.7724)
Beta	0.1905	0.0438
	(0)	(0.316)

View Table

Table 15

Kolmogorov Smirnov Test for FII

	Before Matching	After Matching
Size	0.5716	0.048
	(0)	(0.2194)
Book-to-Price	0.3061	0.0303
	(0)	(0.7724)
Beta	0.1905	0.0438
	(0)	(0.316)

A similar analysis for the firms with high DII investment (but low FII investment) is shown in Table 16 and 17. In the raw data, there are serious problems of match balance, but after matching, the standardised differences are near zero, and the null in the K-S test is not rejected.

Table 16

Standardised Difference for DII

Table 16

Standardised Difference for DII

	Before Matching	After Matching
Size	0.50	0.03
Book-to-Price	−0.07	0.00
Beta	0.13	−0.01

View Table

Table 16

Standardised Difference for DII

	Before Matching	After Matching
Size	0.50	0.03
Book-to-Price	−0.07	0.00
Beta	0.13	−0.01

Table 17

Kolmogorov Smirnov Test for DII

Table 17

Kolmogorov Smirnov Test for DII

	Before Matching	After Matching
Size	0.2342	0.0337
	(0)	(0.1031)
Book-to-Price	0.0513	0.0191
	(0)	(0.7249)
Beta	0.0973	0.0257
	(0)	(0.3566)

View Table

Table 17

Kolmogorov Smirnov Test for DII

	Before Matching	After Matching
Size	0.2342	0.0337
	(0)	(0.1031)
Book-to-Price	0.0513	0.0191
	(0)	(0.7249)
Beta	0.0973	0.0257
	(0)	(0.3566)

For both kinds of institutional investors, this analysis persuades us that the matching process has resulted in a sound design. That is, we will be comparing a firm chosen by an FII or a DII against one that was not chosen by either, while ensuring that there are no systematic differences in size, B/P and β. As emphasised earlier, this ensures that we are focused on the security analysis by FIIs and DIIs, without being clouded by their asset allocation strategies.

6.2 Firms that got high FII but low DII investment

We now analyse the future outcomes for firms that got high FII investment, but low DII investment. These results are shown in Table 18. While conventional OLS results with all data are also shown, we focus on the quasi-experimental design obtained through matching.

Table 18

Outcomes for firms chosen by FIIs but not DIIs

The dataset is a series of matched pairs, where D = 1 is a firm with high FII investment (but low DII investment), and D = 0 is a matched partner which got neither FII nor DII investment. Matching has been done on size, B/P and β, and there is high quality match balance. OLS estimates for y_i,t = β₀ + γD_i,t−j + e_i,t are estimated for various outcomes of interest y, for values of j, and the estimated $\hat{\gamma }$ is reported in each case. Clustered robust standard errors are reported. While the main focus is on the matching-based estimates, conventional OLS estimates using unfiltered data are also reported.

As an example, consider an outcome of interest: log total assets. The matching based estimate shows that the firms chosen by FIIs have a change in log total assets over a three year horizon that is larger than that observed for controls by 0.15, with a standard error of 0.047.

Table 18

Outcomes for firms chosen by FIIs but not DIIs

The dataset is a series of matched pairs, where D = 1 is a firm with high FII investment (but low DII investment), and D = 0 is a matched partner which got neither FII nor DII investment. Matching has been done on size, B/P and β, and there is high quality match balance. OLS estimates for y_i,t = β₀ + γD_i,t−j + e_i,t are estimated for various outcomes of interest y, for values of j, and the estimated $\hat{\gamma }$ is reported in each case. Clustered robust standard errors are reported. While the main focus is on the matching-based estimates, conventional OLS estimates using unfiltered data are also reported.

As an example, consider an outcome of interest: log total assets. The matching based estimate shows that the firms chosen by FIIs have a change in log total assets over a three year horizon that is larger than that observed for controls by 0.15, with a standard error of 0.047.

Log Gross Fixed Assets
	OLS	Matching
Xt − Xt−2	0.08 (0.033) *	0.11 (0.036) **
Xt − Xt−1	0.04 (0.02) *	0.06 (0.021) **
Xt+1 − Xt	0.06 (0.021) **	0.06 (0.018) **
Xt+2 − Xt	0.08 (0.033) *	0.14 (0.039) ***
Xt+3 − Xt	0.12 (0.044) **	0.23 (0.057) ***

Log Employment
	OLS	Matching
Xt − Xt−2	0.17 (0.044) ***	0.1 (0.038) *
Xt − Xt−1	0.11 (0.025) ***	0.05 (0.022) *
Xt+1 − Xt	0.02 (0.024)	0.03 (0.022)
Xt+2 − Xt	0.07 (0.047)	0.07 (0.044)
Xt+3 − Xt	0.07 (0.067)	0.11 (0.069)

Cobb Douglas Productivity Growth
	OLS	Matching
Xt − Xt−2	0.05 (0.033)	−0.06 (0.027) *
Xt − Xt−1	0.09 (0.056)	−0.11 (0.045) *
Xt+1 − Xt	0.01 (0.028)	−0.03 (0.03)
Xt+2 − Xt	0.05 (0.049)	−0.09 (0.055)
Xt+3 − Xt	0.07 (0.069)	−0.05 (0.075)

Log Adjusted Closing Price
	OLS	Matching
Xt − Xt−2	0.21 (0.092) *	−0.05 (0.057)
Xt − Xt−1	0.19 (0.06) **	−0.01 (0.035)
Xt+1 − Xt	0.01 (0.07)	−0.04 (0.038)
Xt+2 − Xt	−0.08 (0.106)	−0.11 (0.069)
Xt+3 − Xt	−0.09 (0.128)	−0.12 (0.103)

Log Total Assets
	OLS	Matching
Xt − Xt−2	0.15 (0.036) ***	0.1 (0.027) ***
Xt − Xt−1	0.1 (0.02) ***	0.05 (0.014) ***
Xt+1 − Xt	0.08 (0.019) ***	0.05 (0.014) **
Xt+2 − Xt	0.11 (0.031) ***	0.1 (0.029) **
Xt+3 − Xt	0.13 (0.043) **	0.15 (0.047) **

Log Sales
	OLS	Matching
Xt − Xt−2	0.04 (0.033)	0.03 (0.042)
Xt − Xt−1	0.03 (0.018) .	0.02 (0.024)
Xt+1 − Xt	0.06 (0.021) **	0.03 (0.025)
Xt+2 − Xt	0.11 (0.041) **	0.04 (0.046)
Xt+3 − Xt	0.2 (0.064) **	0.16 (0.07) *

TFP (Log LP estimate)
	OLS	Matching
Xt − Xt−2	−0.01 (0.016)	0 (0.01)
Xt − Xt−1	0 (0.008)	0 (0.006)
Xt+1 − Xt	−0.01 (0.01)	−0.01 (0.007) .
Xt+2 − Xt	0.01 (0.012)	−0.01 (0.012)
Xt+3 − Xt	0.01 (0.013)	0 (0.016)

View Table

Table 18

Outcomes for firms chosen by FIIs but not DIIs

The dataset is a series of matched pairs, where D = 1 is a firm with high FII investment (but low DII investment), and D = 0 is a matched partner which got neither FII nor DII investment. Matching has been done on size, B/P and β, and there is high quality match balance. OLS estimates for y_i,t = β₀ + γD_i,t−j + e_i,t are estimated for various outcomes of interest y, for values of j, and the estimated $\hat{\gamma }$ is reported in each case. Clustered robust standard errors are reported. While the main focus is on the matching-based estimates, conventional OLS estimates using unfiltered data are also reported.

As an example, consider an outcome of interest: log total assets. The matching based estimate shows that the firms chosen by FIIs have a change in log total assets over a three year horizon that is larger than that observed for controls by 0.15, with a standard error of 0.047.

Log Gross Fixed Assets
	OLS	Matching
Xt − Xt−2	0.08 (0.033) *	0.11 (0.036) **
Xt − Xt−1	0.04 (0.02) *	0.06 (0.021) **
Xt+1 − Xt	0.06 (0.021) **	0.06 (0.018) **
Xt+2 − Xt	0.08 (0.033) *	0.14 (0.039) ***
Xt+3 − Xt	0.12 (0.044) **	0.23 (0.057) ***

Log Employment
	OLS	Matching
Xt − Xt−2	0.17 (0.044) ***	0.1 (0.038) *
Xt − Xt−1	0.11 (0.025) ***	0.05 (0.022) *
Xt+1 − Xt	0.02 (0.024)	0.03 (0.022)
Xt+2 − Xt	0.07 (0.047)	0.07 (0.044)
Xt+3 − Xt	0.07 (0.067)	0.11 (0.069)

Cobb Douglas Productivity Growth
	OLS	Matching
Xt − Xt−2	0.05 (0.033)	−0.06 (0.027) *
Xt − Xt−1	0.09 (0.056)	−0.11 (0.045) *
Xt+1 − Xt	0.01 (0.028)	−0.03 (0.03)
Xt+2 − Xt	0.05 (0.049)	−0.09 (0.055)
Xt+3 − Xt	0.07 (0.069)	−0.05 (0.075)

Log Adjusted Closing Price
	OLS	Matching
Xt − Xt−2	0.21 (0.092) *	−0.05 (0.057)
Xt − Xt−1	0.19 (0.06) **	−0.01 (0.035)
Xt+1 − Xt	0.01 (0.07)	−0.04 (0.038)
Xt+2 − Xt	−0.08 (0.106)	−0.11 (0.069)
Xt+3 − Xt	−0.09 (0.128)	−0.12 (0.103)

Log Total Assets
	OLS	Matching
Xt − Xt−2	0.15 (0.036) ***	0.1 (0.027) ***
Xt − Xt−1	0.1 (0.02) ***	0.05 (0.014) ***
Xt+1 − Xt	0.08 (0.019) ***	0.05 (0.014) **
Xt+2 − Xt	0.11 (0.031) ***	0.1 (0.029) **
Xt+3 − Xt	0.13 (0.043) **	0.15 (0.047) **

Log Sales
	OLS	Matching
Xt − Xt−2	0.04 (0.033)	0.03 (0.042)
Xt − Xt−1	0.03 (0.018) .	0.02 (0.024)
Xt+1 − Xt	0.06 (0.021) **	0.03 (0.025)
Xt+2 − Xt	0.11 (0.041) **	0.04 (0.046)
Xt+3 − Xt	0.2 (0.064) **	0.16 (0.07) *

TFP (Log LP estimate)
	OLS	Matching
Xt − Xt−2	−0.01 (0.016)	0 (0.01)
Xt − Xt−1	0 (0.008)	0 (0.006)
Xt+1 − Xt	−0.01 (0.01)	−0.01 (0.007) .
Xt+2 − Xt	0.01 (0.012)	−0.01 (0.012)
Xt+3 − Xt	0.01 (0.013)	0 (0.016)

The first outcome variable that we analyse is log gross fixed assets. When we look back in time, we see that the firms where D = 1, i.e. the firms with high FII investment (but not high DII investment) got faster growth in fixed assets in the one and two years prior to observation date. In other words, FIIs appear to be choosing firms which have experienced high growth in fixed assets. Looking into the future, the firms chosen by FIIs had a change in log fixed assets that was larger than the control by 0.06 on a horizon of one year, 0.14 on a horizon of two years and 0.23 on a horizon of three years. All these differences are strongly statistically significant. This suggests that the firms chosen by FIIs increased their fixed assets strongly in the period following selection by FIIs. This could either reflect selectivity by FIIs (i.e. FIIs chose firms which were likely to grow well) and it could reflect a causal effect as well (the purchase of shares on the secondary market by FIIs impacted upon the growth of the firm).

Similar results are obtained for log total assets. FIIs chose firms where the balance sheet grew faster in the preceding one and two years. After the measurement date, the firms chosen by FIIs had a change in log total assets that was larger than the control by 0.05 on a horizon of one year, 0.1 on a horizon of two years and 0.15 on a horizon of three years. All these differences were strongly statistically significant.

Turning to employment growth, the firms chosen by FIIs had weakly superior employment growth in the years prior to measurement date. After the measurement date, their employment growth was only slightly greater than the control. The firms chosen by FIIs thus appear to have pursued capital-intensive growth strategies in the years after measurement date.

Despite strong increases in capital, and slight increases in employment, on a horizon of one and two years after measurement date, output growth by selected firms was not significantly greater than the control.

A simple productivity measure as described above yields striking and negative estimates. The firms chosen by FIIs were inferior compared with the control in the two years prior to measurement date. This reflects the combination of high growth of capital but weak growth in output.

For manufacturing firms, we are able to compare TFP using Levinsohn-Petrin estimates. This shows no statistically significant difference between the firms chosen by FIIs and the controls.

Finally, we look at stock market returns. Firms chosen by FIIs show a signicantly lesser return than the controls over a two and three year horizon.

To summarise, these results suggest that the firms chosen by FIIs are on a trajectory of capital deepening. In the period after measurement date, there is strong growth of capital when compared with the control, which could either reflect forecasts of high growth by the FII, or a causal effect of the purchase of shares by FIIs. There is weak evidence of increased employment. However, the impact of output, productivity and stock market returns is surprisingly weak. On horizons of one, two and three years, it is hard to suggest that the firms chosen by FIIs fared better than the controls, either in terms of returns or in terms of growth in output and employment.

6.3 Firms that got high DII but low FII investment

We now turn to the firms chosen by DIIs but not FIIs, where results are in Table 19. In the years prior to the measurement date, the firms selected by DIIs had lower growth in fixed assets and in total assets. In the years after measurement date, their growth of capital is not statistically significantly different from the control. This may be interpreted as investing in distressed companies.

Table 19

Outcomes for firms chosen by DIIs but not FIIs

The dataset is a series of matched pairs, where D = 1 is a firm with high DII investment (but low FII investment), and D = 0 is a matched partner which got neither FII nor DII investment. Matching has been done on size, B/P and β, and there is high quality match balance. OLS estimates for y_i,t = β₀ + γD_i,t−j + e_i,t are estimated for various outcomes of interest y, for values of j, and the estimated $\hat{\gamma }$ is reported in each case. Clustered robust standard errors are reported. While the main focus is on the matching-based estimates, conventional OLS estimates using unfiltered data are also reported.

As an example, consider an outcome of interest: log total assets. The matching based estimate shows that the firms chosen by DIIs have a change in log total assets two years prior to measurement date that is larger than that observed for controls by -0.06, with a standard error of 0.015.

Table 19

Outcomes for firms chosen by DIIs but not FIIs

The dataset is a series of matched pairs, where D = 1 is a firm with high DII investment (but low FII investment), and D = 0 is a matched partner which got neither FII nor DII investment. Matching has been done on size, B/P and β, and there is high quality match balance. OLS estimates for y_i,t = β₀ + γD_i,t−j + e_i,t are estimated for various outcomes of interest y, for values of j, and the estimated $\hat{\gamma }$ is reported in each case. Clustered robust standard errors are reported. While the main focus is on the matching-based estimates, conventional OLS estimates using unfiltered data are also reported.

As an example, consider an outcome of interest: log total assets. The matching based estimate shows that the firms chosen by DIIs have a change in log total assets two years prior to measurement date that is larger than that observed for controls by -0.06, with a standard error of 0.015.

Log Gross Fixed Assets
	OLS	Matching
Xt − Xt−2	−0.1 (0.02) ***	−0.05 (0.02) *
Xt − Xt−1	−0.05 (0.011) ***	−0.02 (0.01) .
Xt+1 − Xt	−0.03 (0.01) *	−0.01 (0.012)
Xt+2 − Xt	−0.05 (0.02) **	−0.02 (0.023)
Xt+3 − Xt	−0.05 (0.031)	−0.04 (0.033)

Log Employment
	OLS	Matching
Xt − Xt−2	−0.06 (0.024) **	−0.02 (0.021)
Xt − Xt−1	−0.03 (0.013) *	0 (0.013)
Xt+1 − Xt	−0.03 (0.015) .	0.01 (0.013)
Xt+2 − Xt	−0.03 (0.026)	−0.02 (0.024)
Xt+3 − Xt	−0.01 (0.036)	−0.04 (0.035)

Cobb Douglas Productivity Growth
	OLS	Matching
Xt − Xt−2	0.02 (0.018)	0.02 (0.016)
Xt − Xt−1	0.05 (0.03)	0.04 (0.026)
Xt+1 − Xt	0.03 (0.018)	0 (0.017)
Xt+2 − Xt	0.09 (0.031) **	0.03 (0.031)
Xt+3 − Xt	0.13 (0.043) **	0.08 (0.042) *

Log Adjusted Closing Price
	OLS	Matching
Xt − Xt−2	0.05 (0.052)	0.03 (0.035)
Xt − Xt−1	0.06 (0.03) .	0.04 (0.021) .
Xt+1 − Xt	0.17 (0.034) ***	0.07 (0.023) **
Xt+2 − Xt	0.25 (0.053) ***	0.12 (0.038) **
Xt+3 − Xt	0.26 (0.066) ***	0.18 (0.057) **

Log Total Assets
	OLS	Matching
Xt − Xt−2	−0.1 (0.018) ***	−0.06 (0.015) ***
Xt − Xt−1	−0.04 (0.01) ***	−0.01 (0.008) .
Xt+1 − Xt	−0.01 (0.01)	−0.01 (0.009)
Xt+2 − Xt	−0.02 (0.018)	−0.02 (0.017)
Xt+3 − Xt	−0.01 (0.026)	−0.02 (0.025)

Log Sales
	OLS	Matching
Xt − Xt−2	−0.05 (0.021) *	0 (0.025)
Xt − Xt−1	−0.02 (0.012)	0.01 (0.015)
Xt+1 − Xt	0.01 (0.015)	0 (0.017)
Xt+2 − Xt	0.03 (0.028)	0.01 (0.032)
Xt+3 − Xt	0.05 (0.04)	0.06 (0.044)

TFP (Log LP estimate)
	OLS	Matching
Xt − Xt−2	0 (0.007)	0 (0.006)
Xt − Xt−1	0 (0.005)	0 (0.004)
Xt+1 − Xt	0.01 (0.005) .	0 (0.004)
Xt+2 − Xt	0.02 (0.006) *	0.01 (0.006)
Xt+3 − Xt	0.01 (0.008) .	0.01 (0.01)

View Table

Table 19

Outcomes for firms chosen by DIIs but not FIIs

The dataset is a series of matched pairs, where D = 1 is a firm with high DII investment (but low FII investment), and D = 0 is a matched partner which got neither FII nor DII investment. Matching has been done on size, B/P and β, and there is high quality match balance. OLS estimates for y_i,t = β₀ + γD_i,t−j + e_i,t are estimated for various outcomes of interest y, for values of j, and the estimated $\hat{\gamma }$ is reported in each case. Clustered robust standard errors are reported. While the main focus is on the matching-based estimates, conventional OLS estimates using unfiltered data are also reported.

As an example, consider an outcome of interest: log total assets. The matching based estimate shows that the firms chosen by DIIs have a change in log total assets two years prior to measurement date that is larger than that observed for controls by -0.06, with a standard error of 0.015.

Log Gross Fixed Assets
	OLS	Matching
Xt − Xt−2	−0.1 (0.02) ***	−0.05 (0.02) *
Xt − Xt−1	−0.05 (0.011) ***	−0.02 (0.01) .
Xt+1 − Xt	−0.03 (0.01) *	−0.01 (0.012)
Xt+2 − Xt	−0.05 (0.02) **	−0.02 (0.023)
Xt+3 − Xt	−0.05 (0.031)	−0.04 (0.033)

Log Employment
	OLS	Matching
Xt − Xt−2	−0.06 (0.024) **	−0.02 (0.021)
Xt − Xt−1	−0.03 (0.013) *	0 (0.013)
Xt+1 − Xt	−0.03 (0.015) .	0.01 (0.013)
Xt+2 − Xt	−0.03 (0.026)	−0.02 (0.024)
Xt+3 − Xt	−0.01 (0.036)	−0.04 (0.035)

Cobb Douglas Productivity Growth
	OLS	Matching
Xt − Xt−2	0.02 (0.018)	0.02 (0.016)
Xt − Xt−1	0.05 (0.03)	0.04 (0.026)
Xt+1 − Xt	0.03 (0.018)	0 (0.017)
Xt+2 − Xt	0.09 (0.031) **	0.03 (0.031)
Xt+3 − Xt	0.13 (0.043) **	0.08 (0.042) *

Log Adjusted Closing Price
	OLS	Matching
Xt − Xt−2	0.05 (0.052)	0.03 (0.035)
Xt − Xt−1	0.06 (0.03) .	0.04 (0.021) .
Xt+1 − Xt	0.17 (0.034) ***	0.07 (0.023) **
Xt+2 − Xt	0.25 (0.053) ***	0.12 (0.038) **
Xt+3 − Xt	0.26 (0.066) ***	0.18 (0.057) **

Log Total Assets
	OLS	Matching
Xt − Xt−2	−0.1 (0.018) ***	−0.06 (0.015) ***
Xt − Xt−1	−0.04 (0.01) ***	−0.01 (0.008) .
Xt+1 − Xt	−0.01 (0.01)	−0.01 (0.009)
Xt+2 − Xt	−0.02 (0.018)	−0.02 (0.017)
Xt+3 − Xt	−0.01 (0.026)	−0.02 (0.025)

Log Sales
	OLS	Matching
Xt − Xt−2	−0.05 (0.021) *	0 (0.025)
Xt − Xt−1	−0.02 (0.012)	0.01 (0.015)
Xt+1 − Xt	0.01 (0.015)	0 (0.017)
Xt+2 − Xt	0.03 (0.028)	0.01 (0.032)
Xt+3 − Xt	0.05 (0.04)	0.06 (0.044)

TFP (Log LP estimate)
	OLS	Matching
Xt − Xt−2	0 (0.007)	0 (0.006)
Xt − Xt−1	0 (0.005)	0 (0.004)
Xt+1 − Xt	0.01 (0.005) .	0 (0.004)
Xt+2 − Xt	0.02 (0.006) *	0.01 (0.006)
Xt+3 − Xt	0.01 (0.008) .	0.01 (0.01)

With employment and output, there is no difference between the firms chosen by DIIs and the controls, in the period before measurement date. But in terms of output growth, firms chosen by DIIs experience higher growth at a three year horizon.

When we examine the simple measure of productivity growth on a three year horizon, the firms chosen by DIIs outperform the controls by a factor of 0.08, which is statistically significant at a 95 per cent level. However, when this is repeated within manufacturing firms only using the Levinsohn-Petrin measure of TFP, the differences in productivity is 0.01.

The most interesting results are found with stock market returns. On horizons of one, two and three years, the firms chosen by DIIs outperform the controls. The superior returns are economically significant: 7 per cent on a one year horizon (with a standard error of 2.3 percentage points), 12 per cent on a two year horizon (with a standard error of 3.8 percentage points) and 18 per cent on a three year horizon (with a standard error of 5.7 percentage points).

The firms chosen by DIIs yield superior stock market returns when compared with controls, while the firms chosen by FIIs do not. This suggests that DIIs possess a valuable investment technology while FIIs do not. While the firms that FIIs invest in experience exuberant growth, there are concerns about productivity, and superior stock market returns are not obtained. In contrast, DIIs appear to get involved in firms that are experiencing difficulties. However, there is some evidence of gains in productivity and strong evidence about superior stock market returns.